Great and less great SSDs for Ethereum nodes

HallOfBlame.md

Overview

Syncing an Ethereum node is largely reliant on IOPS, I/O Per Second. Budget SSDs will struggle to an extent, and some won't be able to sync at all.

This document aims to snapshot some known good and known bad models.

For size, 4TB or 2TB come recommended as of mid 2024. A 2TB drive should last an Ethereum full node until late 2025 or thereabouts, with crystal ball uncertainty.

High-level, QLC and DRAMless are far slower than "mainstream" SSDs. QLC has lower endurance as well. Any savings will be gone when the drive fails early and needs to be replaced.

Other than a slow SSD model, these are things that can slow IOPS down:

• Heat. Check with smartctl -x; the SSD should be below 50C so it does not throttle.
• TRIM not being allowed. This can happen with some hardware RAID controllers, as well as on macOS with non-Apple SSDs
• ZFS
• RAID5/6 - write amplification is no joke
• On SATA, the controller in UEFI/BIOS set to anything other than AHCI. Set it to AHCI for good performance.

If you haven't already, do turn off atime on your DB volume, it'll increase SSD lifetime and speed things up a little bit.

Some users have reported that NUC instability with certain drives can be cured by adding nvme_core.default_ps_max_latency_us=0 to their GRUB_CMDLINE_LINUX kernel parameters via sudo nano /etc/default/grub and sudo update-grub. This keeps the drive from entering powersave states by itself.

The drive lists are ordered by interface and alphabetically by vendor name, not by preference. The lists are not exhaustive at all. @mwpastore linked a filterable spreadsheet in comments that has a far greater variety of drives and their characteristics. Filter it by DRAM yes, NAND Type TLC, Form Factor M.2, and desired capacity.

The Good

"Mainstream" and "Performance" drive models that can sync mainnet execution layer clients in a reasonable amount of time. Use M.2 NVMe if your machine supports it.

Note that in some cases older "Performance" PCIe 4 drives can be bought at a lower price than a PCIe 3 "Mainstream" drive - shop around.

• Often on sale: Samsung 970 EVO Plus, SK Hynix P31 Gold, Teamgroup MP34
• Higher endurance (TBW) than most: Seagate Firecuda 530, WD Red SN700
• Lowest power draw: SK Hynix P31 Gold - great choice for Rock5 B and other low-power devices
• 4TB available: Noted beside each drive, not an exhaustive list

We've started crowd-sourcing some IOPS numbers. If you want to join the fun, run fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75 and give us the read and write IOPS. Don't forget to rm test after.

Hardware

M.2 NVMe "Mainstream" - TLC, DRAM, PCIe 3

• Any data center/enterprise NVMe SSD
• AData XPG Gammix S11/SX8200 Pro. Several hardware revisions. It's slower than some QLC drives. 68k/22k r/w IOPS
• AData XPG Gammix S50 Lite
• HP EX950
• Mushkin Pilot-E
• Samsung 970 EVO Plus 2TB, pre-rework (firmware 2B2QEXM7). 140k/46k r/w IOPS
• Samsung 970 EVO Plus 2TB, post-rework (firmware 3B2QEXM7 or 4B2QEXM7). In testing this syncs just as quickly as the pre-rework drive
• SK Hynix P31 Gold
• Team MP34, 2TB/4TB between 94k/31k and 118k/39k r/w IOPS
• WD Black SN750 (but not SN750 SE)
• WD Red SN700, 2TB/4TB 141k/47k r/w IOPS

M.2 NVMe "Performance" - TLC, DRAM, PCIe 4 or 5

• Any data center/enterprise NVMe SSD
• ADATA XPG Gammix S70, 2TB/4TB, 272k/91k r/w IOPS
• Corsair Force MP600 Pro and variants (but not "MP600 Core XT"), 2TB/4TB/8TB
• Crucial P5 Plus
• Kingston KC2000
• Kingston KC3000, 2TB/4TB, 377k/126k r/w IOPS
• Kingston Fury Renegade, 2TB/4TB
• Mushkin Redline Vortex (but not LX), 2TB/4TB
• Sabrent Rocket 4 Plus, 2TB/4TB/8TB
• Sabrent Rocket 4 Plus-G, 2TB/4TB
• Samsung 980 Pro (not 980) - a firmware update to 5B2QGXA7 is necessary to keep them from dying, if they are firmware 3B2QGXA7. Samsung's boot Linux is a bit broken, you may want to flash from your own Linux.
• Samsung 990 Pro, 124k/41k r/w IOPS 2TB/4TB - there are reports of 990 Pro rapidly losing health. A firmware update to 1B2QJXD7 is meant to stop the rapid degradation, but won't reverse any that happened on earlier firmware.
• Seagate Firecuda 530, 2TB/4TB 218k/73k r/w IOPS
• SK Hynix P41 Platinum / Solidigm P44 Pro, 99k/33k r/w IOPS
• Teamgroup MP44, 2TB/4TB/8TB 105k/35k r/w IOPS - caution that this is DRAMless and uses a Host Memory Buffer (HMB), yet appears to perform fine.
• WD Black SN850
• WD Black SN850X, 2TB/4TB 101k/33k r/w IOPS

Cloud

• Any baremetal/dedicated server service
• AWS i3en.2xlarge
• AWS gp3 w/ >=10k IOPS provisioned and an m7i/a.xlarge

The Bad

These "Budget" drive models are reportedly too slow to sync (all) mainnet execution layer clients.

Hardware

• AData S40G/SX8100 4TB, QLC - the 2TB model is TLC and should be fine; 4TB is reportedly too slow
• Crucial P1, QLC - users report it can't sync Nethermind
• Crucial P2 and P3 (Plus), QLC and DRAMless - users report it can't sync Nethermind, 27k/9k r/w IOPS
• Kingston NV1 - probably QLC and DRAMless and thus too slow on 2TB, but could be "anything" as Kingston do not guarantee specific components.
• Kingston NV2 - like NV1 no guaranteed components
• WD Green SN350, QLC and DRAMless
• Anything both QLC and DRAMless will likely not be able to sync at all or not be able to consistently keep up with "chain head"
• Crucial BX500 SATA, HP S650 SATA, probably most SATA budget drives
• Samsung 980, DRAMless - unsure, this may belong in "Ugly". If you have one and can say for sure, please come to ethstaker Discord.
• Samsung T7 USB, even with current firmware

The Ugly

"Budget" drive models that reportedly can sync mainnet execution layer clients, if slowly.

Note that QLC drives usually have a markedly lower TBW than TLC, and will fail earlier.

Hardware

• Corsair MP400, QLC
• Inland Professional 3D NAND, QLC
• Intel 660p, QLC. It's faster than some "mainstream" drives. 98k/33k r/w IOPS
• Seagata Barracuda Q5, QLC
• WD Black SN770, DRAMless
• Samsung 870 QVO SATA, QLC

2.5" SATA "Mainstream" - TLC, DRAM

• These have been moved to "ugly" because there are user reports that only Nimbus/Geth will now sync on SATA, and even that takes 3 days. It looks like after Dencun, NVMe is squarely the way to go.
• Any data center/enterprise SATA SSD
• Crucial MX500 SATA, 46k/15k r/w IOPS
• Samsung 860 EVO SATA, 55k/18k r/w IOPS
• Samsung 870 EVO SATA, 63k/20k r/w IOPS
• WD Blue 3D NAND SATA

Cloud

• Contabo SSD - reportedly able to sync Geth 1.13.0 and Nethermind, if slowly
• Netcup VPS Servers - reportedly able to sync Geth 1.13.0 and Nethermind, if slowly
• Contabo NVMe - fast enough but not enough space. 800 GiB is not sufficient.
• Netcup RS Servers. Reportedly fast enough to sync Nethermind or Geth; still no speed demon.

I just hit a new personal best, syncing a mainnet full node (not archive) from scratch using Erigon and Lighthouse in roughly 48 hours after downloading snapshots.

• Erigon 2.30.0: with --externalcl --prune=htcr
• Lighthouse 3.2.1: with checkpoint sync against my primary Lighthouse instance, without --sync-historic-states
• CPU: i5-13600K (P-cores only, stock settings)
• RAM: 32GB DDR4-3200 CL16 CR2 1:1 (Gear 1), no swap
• OS: Ubuntu Server 22.04.1 LTS
• Storage:
  • Erigon:
    • snapshots: Samsung 970 EVO+ 2TB
    • temp: Samsung 970 EVO+ 2TB
    • everything else: Intel Optane 905p 960GB
  • Lighthouse: Samsung 970 EVO+ 2TB
  • LVM2, all LVs are linear and formatted with XFS

I think the high IPC of these Raptor Cove P-cores helped a ton, and the 905p was able to keep up with the sustained, high, random read IOPS at low queue depths (up to 11K) toward the end of the initial sync. It was still hitting 36–42 blk/s at block 159* (vs ~20 at the same point on my 5950X and SK hynix P41 2TB). The 905p also seems to be able to transition more quickly between the batch assembly and commit phases—although not much moreso than a high-end Gen4 NAND flash drive like I expected.

What I'll do now is move Erigon's main LV to the 970 EVO+ and repeat the exact same sync. That will help me isolate how much of the speedup is from the Optane drive vs. the rest of the hardware. Specifically, per your notes on CPU performance, I'd like to know the difference between syncing with Raptor Cove vs Zen 3 P-cores.

UPDATE: After ~24 hours I'm at block 127* and already dropping below 40 blk/s. So far at least it seems like the 905p did speed things up measurably on otherwise the same hardware, but it remains to be seen how significantly it affected total sync time.

UPDATE 2: After ~48 hours I'm at block 153* and already dropping below 30 blk/s. The system appears to be I/O-bound per PSI; sustained random read IOPS at low queue depths seems to be peaking around 3.7K on the EVO+, or about a third of what I was seeing on the 905p at roughly the same point in the sync process.

UPDATE 3: It finished syncing last night/overnight and unfortunately I didn't check it before I went to bed and wasn't able to capture the exact time in my tmux scrollback buffer. Based on estimated block speed I would guess between 6⅓ and 9½ hours after the 48 hour mark, plus an hour or two for the remaining phases.

So in conclusion if you're largely I/O-bound as you get into blocks from 2021 and later then yes you could benefit from a drive that features high random read IOPS at low queue depths. In late 2022 that translates to an 8–12 hour speedup (48 hours vs 56–60 hours) with low-end 3D XPoint vs high-end TLC NAND on Raptor Cove. Whether it makes sense to spend hundreds and potentially thousands of dollars more on storage for such a speedup is highly questionable.

The next test would be to see if I can achieve a similar speedup with a smaller allocation of the 905p configured as a writeback cache. That would open up the possibility of pairing a relatively-cheap e.g. P1600X with some less expensive storage.

42 hours with a 110GiB slice of the 905p configured as writeback cache! That exceeds my wildest expectations.

I would still like to play around with writeback vs. writethrough modes but I think my question has been answered: Yes, Optane can significantly speed up the last leg of the initial sync once it's no longer CPU-bound. If you're running on Zen 3 or Golden Cove or newer and want to fully-optimize Erigon re-sync performance, I think it's worth considering.

Newegg has Optane P1600X 118GB drives on sale for $76 a pop. This model doesn't have the absolute best peak specs but it should absolutely destroy QD1 random read. At that price point I will be adding one to my node to complement the two MX500 2TB SATA drives I have in there already.

I'll stop spamming this thread but please let me know if you have questions. Cheers and happy validating!

Very cool, thanks for all the testing! When you say writeback cache, what exactly is the setup you're using?

When you say writeback cache, what exactly is the setup you're using?

LVM2. You create a logical volume, tell LVM2 it's a cache volume, then attach it to another LV. Something like this:

lvcreate -L 110G -n cache_1 vgeth /dev/nvme0n1
lvcreate -L 110M -n cache_1_meta vgeth /dev/nvme0n1
lvconvert --type cache-pool --poolmetadata vgeth/cache_1_meta vgeth/cache_1
lvconvert --type cache --cachepool vgeth/cache_1 --cachemode writeback vgeth/erigon-mainnet

There are easier ways to do it; I use this procedure which includes some extra steps that allow me to optionally put the metadata on a different PV than the data, or use a non-linear topology for the data and/or metadata.

Awesome list. At @AvadoDServer we switched to the Kingston Fury Renegade with Heatsink for 2, 4, 6, 10, 12 TB Versions

Question about testing - using this command:
fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
On one drive I'm trouble-shooting this will take 36 hours. I'm unsure of the mechanism of the test - would I need to wait for the test to complete to gain accurate results, or would stopping it after 2 hours give me results that are close enough?

If it takes that long then that’s a very slow drive. You can absolutely just abandon it and mark the drive as slow.

Thanks for compiling this! Feel free to add this SSD to the ugly:

• Silicon Power 2TB SSD 3D NAND A55 SLC Cache Performance Boost SP002TBSS3A55S25 5.3k/1.7k r/w IOPS. This is not enough for Nethermind to sync.

Crucial P3 Plus 4TB NVMe (listed under The Bad above):

the bad:

sudo fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
[sudo] password for e: 
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=357MiB/s,w=118MiB/s][r=91.3k,w=30.1k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1655: Mon Jul  3 01:06:08 2023
  read: IOPS=53.4k, BW=209MiB/s (219MB/s)(113GiB/551963msec)
   bw (  KiB/s): min=68280, max=366212, per=100.00%, avg=213805.88, stdev=26898.10, samples=1103
   iops        : min=17070, max=91553, avg=53451.40, stdev=6724.53, samples=1103
  write: IOPS=17.8k, BW=69.6MiB/s (72.9MB/s)(37.5GiB/551963msec); 0 zone resets
   bw (  KiB/s): min=22768, max=119775, per=100.00%, avg=71257.56, stdev=8954.47, samples=1103
   iops        : min= 5692, max=29943, avg=17814.27, stdev=2238.60, samples=1103
  cpu          : usr=11.09%, sys=45.81%, ctx=12177744, majf=2, minf=13
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=209MiB/s (219MB/s), 209MiB/s-209MiB/s (219MB/s-219MB/s), io=113GiB (121GB), run=551963-551963msec
  WRITE: bw=69.6MiB/s (72.9MB/s), 69.6MiB/s-69.6MiB/s (72.9MB/s-72.9MB/s), io=37.5GiB (40.3GB), run=551963-551963msec

Disk stats (read/write):
  nvme0n1: ios=29480777/9825548, merge=0/110, ticks=28683132/119856, in_queue=28803427, util=100.00%

Is it worth getting a 4TB drive if I'm just starting out? Looking to run a reth node, and I want to future proof.

Reth requires 4TB at present, as it always syncs an archive node. WD Red SN700 and WD Black SN850(X) are good contenders; there are others.

Once Reth can sync a pruned node, 2TB should work for it, as well.

On their docs it mentions a 2TB for syncing the archive node. Has that turned out to be insufficient? I'm planning on using my SSD with an enclosure and connecting to a MacBook. Is there anything cheaper, or is the SN850X worth the price tag?

The Reth node itself will take ~2TB but it will exceed that soon, and if you're running a CL it'll also take some extra space, so it's generally recommended for your total to be over the 2TB limit so you don't run out of space soon.

In my testing Reth takes around 2 TiB, which means it exceeds a 2TB drive (around 1.9 TiB) right away.

How did you manage to get such IOPS for Seagate Firecuda 530? I am testing a 4TB drive and the IOPS I am getting are 142k/47k:

test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=584MiB/s,w=194MiB/s][r=150k,w=49.6k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=2321: Fri Sep  8 16:46:37 2023
  read: IOPS=142k, BW=556MiB/s (583MB/s)(113GiB/207369msec)
   bw (  KiB/s): min=333224, max=927240, per=100.00%, avg=568914.64, stdev=173766.69, samples=414
   iops        : min=83306, max=231810, avg=142228.69, stdev=43441.70, samples=414
  write: IOPS=47.4k, BW=185MiB/s (194MB/s)(37.5GiB/207369msec); 0 zone resets
   bw (  KiB/s): min=110912, max=306112, per=100.00%, avg=189608.24, stdev=57858.32, samples=414
   iops        : min=27728, max=76528, avg=47402.06, stdev=14464.58, samples=414
  cpu          : usr=15.99%, sys=43.29%, ctx=914957, majf=0, minf=6
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=556MiB/s (583MB/s), 556MiB/s-556MiB/s (583MB/s-583MB/s), io=113GiB (121GB), run=207369-207369msec
  WRITE: bw=185MiB/s (194MB/s), 185MiB/s-185MiB/s (194MB/s-194MB/s), io=37.5GiB (40.3GB), run=207369-207369msec

Disk stats (read/write):
  sdg: ios=29453080/9829388, merge=0/275, ticks=7292548/1934115, in_queue=9226688, util=99.93%

How did you manage to get such IOPS for Seagate Firecuda 530? I am testing a 4TB drive and the IOPS I am getting are 142k/47k:

@valo I have the same Firecuda and getting 254k/84.7k, but I have it on PCIe 3.0 x4 slot, so it's expected (~2x less throughput than PCIe 4.0 x4).
for your case I'd double-check what's your slot's throughput, because it looks like it's ~2x slower than PCIe 3.0 x4.

@d347h-eth I am using Windows 11 with WSL2 and I attach the NVMe directly to WSL2 using wsl2 --mount --bare. Turned out that having Memory integrity enabled in Device security -> Core isolation causes the IOPS to drop in half. After disabling that I get similar results to yours. Interesting find for anyone using a windows + wsl2 system:

test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=843MiB/s,w=281MiB/s][r=216k,w=71.9k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=909: Sun Sep 10 23:05:35 2023
  read: IOPS=218k, BW=853MiB/s (895MB/s)(113GiB/134980msec)
   bw (  KiB/s): min=80937, max=1107312, per=100.00%, avg=875172.81, stdev=131967.58, samples=269
   iops        : min=20234, max=276828, avg=218793.17, stdev=32991.93, samples=269
  write: IOPS=72.8k, BW=284MiB/s (298MB/s)(37.5GiB/134980msec); 0 zone resets
   bw (  KiB/s): min=26981, max=370216, per=100.00%, avg=291685.58, stdev=43874.05, samples=269
   iops        : min= 6745, max=92554, avg=72921.35, stdev=10968.53, samples=269
  cpu          : usr=17.81%, sys=47.06%, ctx=1693386, majf=0, minf=7
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=853MiB/s (895MB/s), 853MiB/s-853MiB/s (895MB/s-895MB/s), io=113GiB (121GB), run=134980-134980msec
  WRITE: bw=284MiB/s (298MB/s), 284MiB/s-284MiB/s (298MB/s-298MB/s), io=37.5GiB (40.3GB), run=134980-134980msec

Disk stats (read/write):
  sdf: ios=29465514/9820322, merge=0/26, ticks=6502679/1003158, in_queue=7505932, util=99.66%

Seems like the drive is working in PCIe 4.0 x4 mode. I'll make an additional test from a linux live system and see if there is any difference.

I am using a Sakura Internet VPS server (Japan) and sometimes Validator crashes.
I measured the IOPS and got the following results.
CPU 8 Core/ Memory 16GB / SSD 1600GB VPS Service.
I may need to change my cloud service.(;;)

<journalctl WARN/error log>
Aug 04 20:14:58 ik1-430-47259 lighthouse[93709]: Aug 04 11:14:58.747 WARN Execution engine call failed error: HttpClient(url: http://localhost:8551/, kind: timeout, detail: operation timed out), service: exec
Aug 04 20:14:58 ik1-430-47259 lighthouse[93709]: Aug 04 11:14:58.747 WARN Error whilst processing payload status error: Api { error: HttpClient(url: http://localhost:8551/, kind: timeout, detail: operation timed out) }, service: exec
Aug 04 20:14:58 ik1-430-47259 lighthouse[93709]: Aug 04 11:14:58.747 CRIT Failed to update execution head error: ExecutionForkChoiceUpdateFailed(EngineError(Api { error: HttpClient(url: http://localhost:8551/, kind: timeout, detail: operation timed out) })), service: beacon

test: (groupid=0, jobs=1): err= 0: pid=663286: Mon Sep 18 18:12:07 2023 read: IOPS=4391, BW=17.2MiB/s (18.0MB/s)(113GiB/6715029msec) bw ( KiB/s): min= 2260, max=23072, per=100.00%, avg=17634.97, stdev=1648.47, samples=13385 iops : min= 565, max= 5768, avg=4408.63, stdev=412.13, samples=13385 write: IOPS=1463, BW=5855KiB/s (5996kB/s)(37.5GiB/6715029msec); 0 zone resets bw ( KiB/s): min= 88, max= 7552, per=100.00%, avg=5876.91, stdev=487.16, samples=13386 iops : min= 22, max= 1888, avg=1469.15, stdev=121.81, samples=13386 cpu : usr=0.82%, sys=6.41%, ctx=4757653, majf=0, minf=6 IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0% submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0% issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0 latency : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
READ: bw=17.2MiB/s (18.0MB/s), 17.2MiB/s-17.2MiB/s (18.0MB/s-18.0MB/s), io=113GiB (121GB), run=6715029-6715029msec
WRITE: bw=5855KiB/s (5996kB/s), 5855KiB/s-5855KiB/s (5996kB/s-5996kB/s), io=37.5GiB (40.3GB), run=6715029-6715029msec

Disk stats (read/write):
vda: ios=30713239/9971460, merge=7896/111624, ticks=17920914/413844545, in_queue=431767366, util=77.61%

As a newbie in this field, I can share an observation. Investing in storage hardware manufacturer's stocks might be as advantageous as investing in cryptocurrency. Building the properly optimized rig to handle this data is quite a task, intellectually and monetarily. I am very impressed with the work you people are doing :-) I do have a question. I read that a "full" archive node with all state back to genesys requires more than 12TB. The most capable motherboard I have on hand has four M.2 slots, 1 Gen5 and 3 Gen4. The largest NVMe SSDs I can buy right now are 4TB and some are Gen5 and expensive. If I pop 3 and the Gen4's into the Gen4 slots that would give me 12TB for the full archive, but its not the fastest. This motherboard also has a ton of SATAs. If its not to much to ask, could you answer the question, whether it would be better from a cost/performance perspective to max out the NVMe4s for the archive, use the 5 for the os, or put the archive on the SATAs? Thanks

It’d be best to run a smaller archive node. Both Reth and Erigon can fit an archive into a single 4TB disk, taking just over 2 TiB. Geth is working on similar capability, early 2024 maybe.

Here is a benchmark of Seagate Firecuda 530 4TB on a native Ubuntu 22.04 install. Same hardware as my WSL2 test from above. As you can see the IOPS are about 50% higher, so if you are using WSL2, migrating to native linux should give you quite a bit of IO speedup. Interestingly I can't replicate the IOPS from the gist for this device.

test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
Jobs: 1 (f=1): [m(1)][100.0%][r=1294MiB/s,w=431MiB/s][r=331k,w=110k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=153177: Thu Sep 28 15:15:18 2023
  read: IOPS=331k, BW=1291MiB/s (1354MB/s)(113GiB/89206msec)
   bw (  MiB/s): min= 1053, max= 1327, per=100.00%, avg=1291.95, stdev=35.50, samples=178
   iops        : min=269642, max=339804, avg=330739.91, stdev=9088.15, samples=178
  write: IOPS=110k, BW=430MiB/s (451MB/s)(37.5GiB/89206msec); 0 zone resets
   bw (  KiB/s): min=362144, max=454496, per=100.00%, avg=440922.97, stdev=11996.22, samples=178
   iops        : min=90536, max=113624, avg=110230.74, stdev=2999.06, samples=178
  cpu          : usr=14.33%, sys=71.05%, ctx=4381204, majf=0, minf=8
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=1291MiB/s (1354MB/s), 1291MiB/s-1291MiB/s (1354MB/s-1354MB/s), io=113GiB (121GB), run=89206-89206msec
  WRITE: bw=430MiB/s (451MB/s), 430MiB/s-430MiB/s (451MB/s-451MB/s), io=37.5GiB (40.3GB), run=89206-89206msec

Disk stats (read/write):
  nvme1n1: ios=29473358/9822980, merge=0/17, ticks=4254089/128160, in_queue=4382314, util=99.92%

As of October 2023:

The Contabo VPS SSD (non-NVMe) is fully capable of running an Ethereum 2.0 mainnet Execution + Validator Node.

Geth version >= 1.13.2 or later with PBSS (path-based database storage) and the Lighthouse validator client.

Geth sync was completed in 12 hours and 30 minutes.

Good to know. I assume Nethermind still cannot sync, or can you report that it now can?

Good to know. I assume Nethermind still cannot sync, or can you report that it now can?

ETH network load is quite low nowadays, but it can Nethermind can sync and keep it up with the chain, but I can confirm Nethermind is much more I/O hungry and touch the IOPS limit.
You can ask higher IOPS at Contabo by the support, usually you got it.

Nethermind logs:

 Processed            18326160     |    819.55 ms  |  slot     11,814 ms | Gas gwei: 7.36 .. 7.36 (9.48) .. 60.93
 Processed            18326161     |  1,154.41 ms  |  slot     14,673 ms | Gas gwei: 7.27 .. 7.27 (11.52) .. 230.44
 Processed            18326162     |    667.72 ms  |  slot      9,832 ms | Gas gwei: 7.64 .. 7.65 (11.12) .. 95.42
 Processed            18326163     |    677.76 ms  |  slot     14,253 ms | Gas gwei: 7.50 .. 7.61 (10.95) .. 257.55
 Processed            18326164     |    642.48 ms  |  slot     11,875 ms | Gas gwei: 7.90 .. 7.98 (11.24) .. 115.91

Kingston KC3000 2TB on Ubuntu 22.04:

test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=1482MiB/s,w=493MiB/s][r=379k,w=126k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1513: Fri Dec  1 15:27:59 2023
  read: IOPS=377k, BW=1473MiB/s (1545MB/s)(113GiB/78201msec)
   bw (  MiB/s): min= 1134, max= 1506, per=100.00%, avg=1474.69, stdev=42.25, samples=156
   iops        : min=290478, max=385538, avg=377520.78, stdev=10814.73, samples=156
  write: IOPS=126k, BW=491MiB/s (515MB/s)(37.5GiB/78201msec); 0 zone resets
   bw (  KiB/s): min=386840, max=514416, per=100.00%, avg=503278.41, stdev=14299.46, samples=156
   iops        : min=96710, max=128604, avg=125819.59, stdev=3574.86, samples=156
  cpu          : usr=14.90%, sys=71.19%, ctx=3805582, majf=0, minf=6
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=1473MiB/s (1545MB/s), 1473MiB/s-1473MiB/s (1545MB/s-1545MB/s), io=113GiB (121GB), run=78201-78201msec
  WRITE: bw=491MiB/s (515MB/s), 491MiB/s-491MiB/s (515MB/s-515MB/s), io=37.5GiB (40.3GB), run=78201-78201msec

Disk stats (read/write):
    dm-0: ios=29487639/9827779, merge=0/0, ticks=2155548/89524, in_queue=2245072, util=99.97%, aggrios=29492812/9829478, aggrmerge=29/85, aggrticks=2198992/106382, aggrin_queue=2305387, aggrutil=99.90%
  nvme0n1: ios=29492812/9829478, merge=29/85, ticks=2198992/106382, in_queue=2305387, util=99.90%

TeamGroup MP34 4TB drive on Ubuntu 22.04:

test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=804MiB/s,w=266MiB/s][r=206k,w=68.0k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=16333: Mon Dec 18 04:00:20 2023
  read: IOPS=118k, BW=459MiB/s (482MB/s)(113GiB/250798msec)
   bw (  KiB/s): min=413304, max=854560, per=100.00%, avg=470602.83, stdev=36300.77, samples=501
   iops        : min=103326, max=213640, avg=117650.70, stdev=9075.17, samples=501
  write: IOPS=39.2k, BW=153MiB/s (161MB/s)(37.5GiB/250798msec); 0 zone resets
   bw (  KiB/s): min=138672, max=282072, per=100.00%, avg=156842.56, stdev=12061.92, samples=501
   iops        : min=34668, max=70520, avg=39210.63, stdev=3015.52, samples=501
  cpu          : usr=9.01%, sys=31.13%, ctx=3536466, majf=0, minf=8
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=459MiB/s (482MB/s), 459MiB/s-459MiB/s (482MB/s-482MB/s), io=113GiB (121GB), run=250798-250798msec
  WRITE: bw=153MiB/s (161MB/s), 153MiB/s-153MiB/s (161MB/s-161MB/s), io=37.5GiB (40.3GB), run=250798-250798msec

Disk stats (read/write):
    dm-0: ios=29469140/9821675, merge=0/0, ticks=13703608/1542052, in_queue=15245660, util=100.00%, aggrios=29492330/9829436, aggrmerge=0/54, aggrticks=13745890/1554511, aggrin_queue=15300436, aggrutil=100.00%
  nvme0n1: ios=29492330/9829436, merge=0/54, ticks=13745890/1554511, in_queue=15300436, util=100.00%

Finally got around to troubleshooting the fairly frequent missed attestations I've been plagued with. It would appear something is amiss with my Seagate Firecuda 520. IOPS are reporting significantly worse than other less performant drives I own. Anything I should look into as a next step? Running Ubuntu 22.04.3 LTS

Seagate Firecuda 520 2TB Gen4 X4 NVMe
size: 1863GiB (2TB)
capabilities: gpt-1.00 partitioned partitioned:gpt
configuration: logicalsectorsize=512 sectorsize=512

test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=76.6MiB/s,w=25.7MiB/s][r=19.6k,w=6567 IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=2400068: Wed Dec 27 01:41:32 2023

read: IOPS=23.6k, BW=92.1MiB/s (96.6MB/s)(113GiB/1250418msec)
bw ( KiB/s): min=40880, max=118128, per=100.00%, avg=94485.13, stdev=12039.63, samples=2500
iops : min=10220, max=29532, avg=23621.04, stdev=3009.87, samples=2500

write: IOPS=7860, BW=30.7MiB/s (32.2MB/s)(37.5GiB/1250418msec); 0 zone resets
bw ( KiB/s): min=13752, max=39503, per=100.00%, avg=31489.84, stdev=4019.81, samples=2500
iops : min= 3438, max= 9875, avg=7872.22, stdev=1004.92, samples=2500

cpu : usr=23.99%, sys=65.19%, ctx=8866372, majf=0, minf=9
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
READ: bw=92.1MiB/s (96.6MB/s), 92.1MiB/s-92.1MiB/s (96.6MB/s-96.6MB/s), io=113GiB (121GB), run=1250418-1250418msec
WRITE: bw=30.7MiB/s (32.2MB/s), 30.7MiB/s-30.7MiB/s (32.2MB/s-32.2MB/s), io=37.5GiB (40.3GB), run=1250418-1250418msec

Disk stats (read/write):
dm-0: ios=30274481/9951689, merge=0/0, ticks=6203536/580080, in_queue=6783616, util=100.00%, aggrios=30056705/9880464, aggrmerge=221778/73883, aggrticks=6170229/418768, aggrin_queue=6594545, aggrutil=100.00%
nvme0n1: ios=30056705/9880464, merge=221778/73883, ticks=6170229/418768, in_queue=6594545, util=100.00%

@netzlvl Run a smartctl -x on it and check temperature. If it's in the 60C+ range it's likely throttling. Anything else "in the path" that could impact IO performance? Virtualization, ZFS?

@yorickdowne Thanks for the suggestions. Temps have been hovering around 50-55C. I just retested with case open and fan blowing on the machine bringing temps down to ~40C and still got the same results. This was otherwise a fresh RPL build, no virtualization or anything fancy. Fairly barebones setup. I'm leaning towards formatting and starting fresh but any other suggestions are welcome!

XPG GAMMIX S70 BLADE 4TB uBUNTU 22.04
CJValdez
fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=1038MiB/s,w=346MiB/s][r=266k,w=88.6k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=4869: Sat Jan 27 05:10:48 2024
read: IOPS=272k, BW=1063MiB/s (1115MB/s)(113GiB/108389msec)
bw ( MiB/s): min= 355, max= 1249, per=99.98%, avg=1062.66, stdev=174.61, samples=216
iops : min=90942, max=319828, avg=272039.96, stdev=44698.94, samples=216
write: IOPS=90.7k, BW=354MiB/s (371MB/s)(37.5GiB/108389msec); 0 zone resets
bw ( KiB/s): min=121848, max=427936, per=99.98%, avg=362675.44, stdev=59711.66, samples=216
iops : min=30462, max=106984, avg=90668.86, stdev=14927.91, samples=216
cpu : usr=17.23%, sys=67.11%, ctx=7809250, majf=1, minf=7
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
READ: bw=1063MiB/s (1115MB/s), 1063MiB/s-1063MiB/s (1115MB/s-1115MB/s), io=113GiB (121GB), run=108389-108389msec
WRITE: bw=354MiB/s (371MB/s), 354MiB/s-354MiB/s (371MB/s-371MB/s), io=37.5GiB (40.3GB), run=108389-108389msec

Disk stats (read/write):
nvme0n1: ios=29516035/9832782, merge=179/932, ticks=4976294/179813, in_queue=5156253, util=99.84%

Kingston Fury Renegade 4 TB on Ubuntu 22.04.3 LTS

Jobs: 1 (f=1): [m(1)][100.0%][r=825MiB/s,w=275MiB/s][r=211k,w=70.3k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=2391: Wed Feb  7 22:40:00 2024
  read: IOPS=211k, BW=825MiB/s (865MB/s)(113GiB/139692msec)
   bw (  KiB/s): min=640384, max=857392, per=100.00%, avg=844813.59, stdev=19162.15, samples=279
   iops        : min=160096, max=214348, avg=211203.44, stdev=4790.55, samples=279
  write: IOPS=70.4k, BW=275MiB/s (288MB/s)(37.5GiB/139692msec); 0 zone resets
   bw (  KiB/s): min=213584, max=286904, per=100.00%, avg=281559.23, stdev=6428.86, samples=279
   iops        : min=53396, max=71726, avg=70389.82, stdev=1607.22, samples=279
  cpu          : usr=14.78%, sys=70.64%, ctx=4460659, majf=0, minf=7
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=825MiB/s (865MB/s), 825MiB/s-825MiB/s (865MB/s-865MB/s), io=113GiB (121GB), run=139692-139692msec
  WRITE: bw=275MiB/s (288MB/s), 275MiB/s-275MiB/s (288MB/s-288MB/s), io=37.5GiB (40.3GB), run=139692-139692msec

Disk stats (read/write):
  nvme0n1: ios=29482363/9825979, merge=0/27, ticks=2949353/147884, in_queue=3097249, util=99.97%

Thanks for sharing it with us.

ubuntu 22.04 WD RED 2TB NAS SSD
WDS400T1R0A-68A4W0
Node won't sync

fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=87.9MiB/s,w=29.5MiB/s][r=22.5k,w=7547 IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1263819: Mon Feb 19 11:44:23 2024
  read: IOPS=25.9k, BW=101MiB/s (106MB/s)(113GiB/1138368msec)
   bw (  KiB/s): min=22104, max=161232, per=100.00%, avg=103709.80, stdev=26209.74, samples=2275
   iops        : min= 5526, max=40308, avg=25927.42, stdev=6552.44, samples=2275
  write: IOPS=8634, BW=33.7MiB/s (35.4MB/s)(37.5GiB/1138368msec); 0 zone resets
   bw (  KiB/s): min= 7600, max=54608, per=100.00%, avg=34564.47, stdev=8736.34, samples=2275
   iops        : min= 1900, max=13652, avg=8641.09, stdev=2184.09, samples=2275
  cpu          : usr=6.15%, sys=23.19%, ctx=22865227, majf=0, minf=17
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=101MiB/s (106MB/s), 101MiB/s-101MiB/s (106MB/s-106MB/s), io=113GiB (121GB), run=1138368-1138368msec
  WRITE: bw=33.7MiB/s (35.4MB/s), 33.7MiB/s-33.7MiB/s (35.4MB/s-35.4MB/s), io=37.5GiB (40.3GB), run=1138368-1138368msec

Disk stats (read/write):
  sda: ios=29508728/9828000, merge=2646/272, ticks=60857576/5856658, in_queue=68659770, util=100.00%

Fedora 38 on Qubes R4.1 (virtualized, but none of this multi-layered filesystem on top of filesystem on top of encryption shenanigans)
XFS with reverse mapping
TeamGroup MP44 8TB (HMB) set to 4k sectors

Jobs: 1 (f=1): [m(1)][100.0%][r=415MiB/s,w=137MiB/s][r=106k,w=35.0k IOPS][eta 00m:00s] 
test: (groupid=0, jobs=1): err= 0: pid=1716: Tue Feb 20 09:15:13 2024
  read: IOPS=105k, BW=412MiB/s (432MB/s)(113GiB/279629msec)
   bw (  KiB/s): min=168648, max=757002, per=100.00%, avg=422125.70, stdev=120676.28, samples=559
   iops        : min=42162, max=189250, avg=105531.34, stdev=30169.07, samples=559
  write: IOPS=35.2k, BW=137MiB/s (144MB/s)(37.5GiB/279629msec); 0 zone resets
   bw (  KiB/s): min=55832, max=251358, per=100.00%, avg=140685.69, stdev=40142.70, samples=559
   iops        : min=13958, max=62839, avg=35171.33, stdev=10035.67, samples=559
  cpu          : usr=13.32%, sys=69.85%, ctx=5737782, majf=0, minf=6
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=412MiB/s (432MB/s), 412MiB/s-412MiB/s (432MB/s-432MB/s), io=113GiB (121GB), run=279629-279629msec
  WRITE: bw=137MiB/s (144MB/s), 137MiB/s-137MiB/s (144MB/s-144MB/s), io=37.5GiB (40.3GB), run=279629-279629msec

Disk stats (read/write):
  xvdi: ios=29480750/9825346, merge=0/1, ticks=3170142/447551, in_queue=3617709, util=100.00%

will edit when I've had a chance to see how the node runs on this
reth+nimbus, both in archive mode, syncs and follows the chain just fine

Crucial P3 Plus 4TB in ThinkCentre m75q gen2 (SSD running hot, no room for cooler)
Took 4 weeks to sync with Erigon (archive node), 4TB almost full already, pruning not possible post-sync.
Can barely keep up with chain head.

Jobs: 1 (f=1): [m(1)][100.0%][r=71.5MiB/s,w=24.0MiB/s][r=18.3k,w=6134 IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1236631: Tue Feb 20 10:29:07 2024
  read: IOPS=27.6k, BW=108MiB/s (113MB/s)(113GiB/1069328msec)
   bw (  KiB/s): min=17440, max=244744, per=100.00%, avg=110395.07, stdev=74976.03, samples=2138
   iops        : min= 4360, max=61186, avg=27598.67, stdev=18744.03, samples=2138
  write: IOPS=9192, BW=35.9MiB/s (37.7MB/s)(37.5GiB/1069328msec); 0 zone resets
   bw (  KiB/s): min= 5400, max=81672, per=100.00%, avg=36792.43, stdev=24997.34, samples=2138
   iops        : min= 1350, max=20418, avg=9198.00, stdev=6249.35, samples=2138
  cpu          : usr=3.62%, sys=25.64%, ctx=13387268, majf=1, minf=7
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=108MiB/s (113MB/s), 108MiB/s-108MiB/s (113MB/s-113MB/s), io=113GiB (121GB), run=1069328-1069328msec
  WRITE: bw=35.9MiB/s (37.7MB/s), 35.9MiB/s-35.9MiB/s (37.7MB/s-37.7MB/s), io=37.5GiB (40.3GB), run=1069328-1069328msec

Disk stats (read/write):
    dm-1: ios=29487924/9833619, merge=0/0, ticks=66472928/1215864, in_queue=67688792, util=100.00%, aggrios=29492470/9835159, aggrmerge=0/0, aggrticks=66456848/1206580, aggrin_queue=67663428, aggrutil=100.00%
    dm-0: ios=29492470/9835159, merge=0/0, ticks=66456848/1206580, in_queue=67663428, util=100.00%, aggrios=29492402/9833761, aggrmerge=68/1401, aggrticks=66122896/1097291, aggrin_queue=67223406, aggrutil=100.00%
  nvme0n1: ios=29492402/9833761, merge=68/1401, ticks=66122896/1097291, in_queue=67223406, util=100.00%

WD Red SN700 2000GB NVME

fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=530MiB/s,w=176MiB/s][r=136k,w=45.0k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=22923: Wed Feb 21 13:22:31 2024
  read: IOPS=141k, BW=551MiB/s (578MB/s)(113GiB/208929msec)
   bw (  KiB/s): min=213392, max=846128, per=99.99%, avg=564555.30, stdev=107083.72, samples=417
   iops        : min=53348, max=211532, avg=141138.86, stdev=26770.92, samples=417
  write: IOPS=47.0k, BW=184MiB/s (193MB/s)(37.5GiB/208929msec); 0 zone resets
   bw (  KiB/s): min=72240, max=282960, per=99.99%, avg=188160.02, stdev=35738.54, samples=417
   iops        : min=18060, max=70740, avg=47040.00, stdev=8934.64, samples=417
  cpu          : usr=14.21%, sys=48.19%, ctx=12030604, majf=0, minf=7
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=551MiB/s (578MB/s), 551MiB/s-551MiB/s (578MB/s-578MB/s), io=113GiB (121GB), run=208929-208929msec
  WRITE: bw=184MiB/s (193MB/s), 184MiB/s-184MiB/s (193MB/s-193MB/s), io=37.5GiB (40.3GB), run=208929-208929msec

Disk stats (read/write):
    dm-0: ios=29453038/9816140, merge=0/0, ticks=12182596/107212, in_queue=12289808, util=99.99%, aggrios=29492326/9829312, aggrmerge=0/0, aggrticks=12181730/113359, aggrin_queue=12295224, aggrutil=99.96%
  nvme0n1: ios=29492326/9829312, merge=0/0, ticks=12181730/113359, in_queue=12295224, util=99.96%

@bussyjd Thanks for confirming that SA500 SATA SSD doesn't cut the mustard. Not unexpected. Great results on the SN700 as expected.

@laurenzberger , thanks for taking one for the team and testing a slow model.

And @duncancmt , I've added the MP44. 8TB with decent speed is a new one. Surprising as it's DRAMless.

TEAMGROUP MP34

fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=697MiB/s,w=231MiB/s][r=179k,w=59.0k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=4068: Wed Feb 21 20:05:50 2024
  read: IOPS=94.5k, BW=369MiB/s (387MB/s)(113GiB/312117msec)
   bw (  KiB/s): min=21624, max=801584, per=99.96%, avg=377803.39, stdev=169305.16, samples=623
   iops        : min= 5406, max=200396, avg=94450.87, stdev=42326.30, samples=623
  write: IOPS=31.5k, BW=123MiB/s (129MB/s)(37.5GiB/312117msec); 0 zone resets
   bw (  KiB/s): min= 7320, max=264896, per=99.96%, avg=125916.83, stdev=56412.81, samples=623
   iops        : min= 1830, max=66224, avg=31479.20, stdev=14103.20, samples=623
  cpu          : usr=8.57%, sys=23.25%, ctx=3343348, majf=0, minf=7
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=369MiB/s (387MB/s), 369MiB/s-369MiB/s (387MB/s-387MB/s), io=113GiB (121GB), run=312117-312117msec
  WRITE: bw=123MiB/s (129MB/s), 123MiB/s-123MiB/s (129MB/s-129MB/s), io=37.5GiB (40.3GB), run=312117-312117msec

Disk stats (read/write):
  nvme0n1: ios=29457519/9817778, merge=0/62, ticks=16948691/2236682, in_queue=19185556, util=100.00%

Can i ask, why

• Corsair Force MP600 Pro (but not "Core XT"), 2TB/4TB
  why not MP600 PRO Core XT 8TB ?

For starters because I'm not aware of that model. "MP600 Pro Core XT" may not exist. The spreadsheet has an "MP600 Core XT", that's DRAMless and QLC, and an "MP600 Pro XT", DRAM and TLC. Afaik all "MP600 Pro" models are fine; "MP600 Core XT" is not.

I recently conducted a performance test on the Samsung SSD 990 Pro 4TB model - with heatsink. To my disappointment, the read/write (R/W) speeds were significantly lower than expected, recording at 125k/41.4k IOPS. Given its high DRAM capacity and fast R/W speeds, I expected much better performance.

test: (groupid=0, jobs=1): err= 0: pid=9866: Fri Mar  1 23:07:07 2024
  read: IOPS=124k, BW=486MiB/s (509MB/s)(113GiB/237104msec)
   bw (  KiB/s): min=473160, max=515672, per=100.00%, avg=497642.84, stdev=3501.77, samples=474
   iops        : min=118290, max=128918, avg=124410.73, stdev=875.44, samples=474
  write: IOPS=41.5k, BW=162MiB/s (170MB/s)(37.5GiB/237104msec); 0 zone resets
   bw (  KiB/s): min=158056, max=171448, per=100.00%, avg=165854.41, stdev=1341.47, samples=474
   iops        : min=39514, max=42862, avg=41463.60, stdev=335.37, samples=474
  cpu          : usr=10.13%, sys=76.00%, ctx=8989513, majf=0, minf=10
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=486MiB/s (509MB/s), 486MiB/s-486MiB/s (509MB/s-509MB/s), io=113GiB (121GB), run=237104-237104msec
  WRITE: bw=162MiB/s (170MB/s), 162MiB/s-162MiB/s (170MB/s-170MB/s), io=37.5GiB (40.3GB), run=237104-237104msec

Disk stats (read/write):
  nvme0n1: ios=29487370/9828723, merge=34/827, ticks=1093523/76506, in_queue=1170137, util=99.98%

I have tested this on two systems:

System 1: HP prodesk 600 G3 - Desktop mini
CPU: I5-7500t: 4 cores 4 threads @2.7 GHz - with integrated graphics
RAM: 16GB ddr4 @ 2400MHz
SSD: Samsung SSD 990 Pro 4TB model - with heatsink on a PCIe 3.0 x4 slot

System 2: Custom built pc
CPU: Ryzen 7 5800x
RAM: 32GB ddr4 @ 3200Mhz
GPU: NVIDIA Geforce RTX 3070ti
SSD: Samsung SSD 990 Pro 4TB model - with heatsink on a PCIe 4.0 x4

Things that I made sure to do:

1. Update to the newest firmware using the official Samsung magician software utility tool - This was very important as it was using old firmware which is detrimental to the drives' health.
2. Test on both PCie 3.0 x4 and PCie 4.0 x4 slots. (didn't make a difference btw)
3. Make sure atime was turned off
4. Testing on both Ubuntu server and Ubuntu desktop - 22.04 and 23.10
5. Went into BIOS settings and made sure everything was completely optimized for IOPS
6. Also made sure that it did not thermally throttle by actively cooling and benchmarking with the smartctl -x utility which @yorickdowne recommended.
7. Make sure that the drive was genuine and not some off-market knockoff by registering its serial number with the official Samsung website.
8. The drive also had 0 power ons and power-on hours, so I confirmed for myself that it was brand new.

I have no clue why these results are so bad, this is supposed to be one of (if not the) the best drive on the market. I went through a bunch of troubleshooting but I am starting to believe that this is what the drive is capable of providing in terms of performance. Also since the performance closely resembles the earlier shown results for the Samsung 970 EVO 2tb .

I'm keen to hear from others who own the same SSD model. If you've conducted similar tests, please share your results. I am very interested to know whether this is an isolated case or a common issue with this SSD model.

If anyone has additional troubleshooting tips please let me know, might as well try some stuff while I am at it

Thanks for the extra-diligent tests! Tbh those results aren’t bad at all, they are right in line with what you expect from a good NVMe drive. As you found, PCIe 3 or 4 makes no difference at all for this use case: It’s all about random read/write, not raw bandwidth.

It’s one of the most stressful things you can ask a drive to do, which is why many budget drives struggle, although they are more than fine for desktop and gaming use.

More data points for the Samsung 990 Pro 4TB w/ Heatsink.
noatime set in fstab, nvme_core.default_ps_max_latency_us=0 set in grub
My drive is using the new v8 process, so has the 0B2QJXG7 firmware.
Fresh install of Ubuntu 22.04 LTS, Xeon E-2136, 2x32GB Kingston ECC @ 2666

I was monitoring temp with smartctl -x, and it did get up there at ~55C, but IOPs are looking good at 260k/86.7k Am I reading this right?

$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=999MiB/s,w=332MiB/s][r=256k,w=85.1k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1676: Fri Mar  1 21:13:32 2024
  read: IOPS=260k, BW=1016MiB/s (1065MB/s)(113GiB/113384msec)
   bw (  KiB/s): min=1001280, max=1063096, per=100.00%, avg=1040833.45, stdev=11991.07, samples=226
   iops        : min=250320, max=265774, avg=260208.42, stdev=2997.74, samples=226
  write: IOPS=86.7k, BW=339MiB/s (355MB/s)(37.5GiB/113384msec); 0 zone resets
   bw (  KiB/s): min=332488, max=355800, per=100.00%, avg=346900.35, stdev=4130.15, samples=226
   iops        : min=83122, max=88950, avg=86725.08, stdev=1032.54, samples=226
  cpu          : usr=15.66%, sys=71.73%, ctx=9318105, majf=0, minf=6
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=1016MiB/s (1065MB/s), 1016MiB/s-1016MiB/s (1065MB/s-1065MB/s), io=113GiB (121GB), run=113384-113384msec
  WRITE: bw=339MiB/s (355MB/s), 339MiB/s-339MiB/s (355MB/s-355MB/s), io=37.5GiB (40.3GB), run=113384-113384msec

Disk stats (read/write):
    dm-0: ios=29428508/9808565, merge=0/0, ticks=1220972/95488, in_queue=1316460, util=99.95%, aggrios=29492333/9829587, aggrmerge=0/72, aggrticks=1226083/100143, aggrin_queue=1326263, aggrutil=99.91%
  nvme0n1: ios=29492333/9829587, merge=0/72, ticks=1226083/100143, in_queue=1326263, util=99.91%

More data points for the Samsung 990 Pro 4TB w/ Heatsink. noatime set in fstab, nvme_core.default_ps_max_latency_us=0 set in grub My drive is using the new v8 process, so has the 0B2QJXG7 firmware. Fresh install of Ubuntu 22.04 LTS, Xeon E-2136, 2x32GB Kingston ECC @ 2666

I was monitoring temp with smartctl -x, and it did get up there at ~55C, but IOPs are looking good at 260k/86.7k Am I reading this right?

$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=999MiB/s,w=332MiB/s][r=256k,w=85.1k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1676: Fri Mar  1 21:13:32 2024
  read: IOPS=260k, BW=1016MiB/s (1065MB/s)(113GiB/113384msec)
   bw (  KiB/s): min=1001280, max=1063096, per=100.00%, avg=1040833.45, stdev=11991.07, samples=226
   iops        : min=250320, max=265774, avg=260208.42, stdev=2997.74, samples=226
  write: IOPS=86.7k, BW=339MiB/s (355MB/s)(37.5GiB/113384msec); 0 zone resets
   bw (  KiB/s): min=332488, max=355800, per=100.00%, avg=346900.35, stdev=4130.15, samples=226
   iops        : min=83122, max=88950, avg=86725.08, stdev=1032.54, samples=226
  cpu          : usr=15.66%, sys=71.73%, ctx=9318105, majf=0, minf=6
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=1016MiB/s (1065MB/s), 1016MiB/s-1016MiB/s (1065MB/s-1065MB/s), io=113GiB (121GB), run=113384-113384msec
  WRITE: bw=339MiB/s (355MB/s), 339MiB/s-339MiB/s (355MB/s-355MB/s), io=37.5GiB (40.3GB), run=113384-113384msec

Disk stats (read/write):
    dm-0: ios=29428508/9808565, merge=0/0, ticks=1220972/95488, in_queue=1316460, util=99.95%, aggrios=29492333/9829587, aggrmerge=0/72, aggrticks=1226083/100143, aggrin_queue=1326263, aggrutil=99.91%
  nvme0n1: ios=29492333/9829587, merge=0/72, ticks=1226083/100143, in_queue=1326263, util=99.91%

Thanks for sharing your experience! I recently applied the "nvme_core.default_ps_max_latency_us=0" tweak and noticed a modest improvement in IOPS, achieving an improved 1k/0.5k. Not much, but I'll take it.

Initially, I wasn't aware of the distinctions between the V7 and V8 models, particularly how some V7 versions were impacted by firmware issues leading to accelerated wear. It turns out that not all firmware versions starting with "0B2QJX" are problematic. After some digging and checking my own SSD, I discovered I own the V8 model manufactured in December 2023.

Here's a quick breakdown of the firmware differences:

• V7 Detrimental firmware: 0B2QJXD7 – Known for causing rapid degradation.

• V8 Safe firmware: 0B2QJXG7 – Does not cause rapid degradation.

Notice the difference between the "D" and "G" in the firmware versions?

Before updating the firmware through Samsung Magician, I took a photo for reference in case any issues arose. My SSD originally had the same firmware version you are currently running.

Regrettably, Samsung's website does not offer the older firmware versions for downgrade, only displaying the newer "4B2QJXD7" version, which my SSD is currently running. Reverting to 0B2QJXG7 would be the ultimate final test in determining whether it is a firmware issue or my device's performance is just worse.

Now obviously I mean "Worse" in a relative way, because the results are more than sufficient to to sync an Ethereum node.
I was comparing against the earlier impressive results of the KC3000, the Firecuda and your IOPS results which are definitely much better at 260k/86.7k. I thought that the Samsung SSD due to its later release date and higher DRAM capacity would be a bit better than those other SSD's.

1x SAMSUNG PM963 MZQLW1T9HMJP-00003 U.2
1x SAMSUNG SM963 MZQKW1T9HMJP-00003 U.2
(Similar SSD's one of them is MLC the other is TLC)
Very durable Server SSDs can be obtained cheap in secondmarket
ON RAID 0 BTRFS

fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.33
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=0): [f(1)][100.0%][r=219MiB/s,w=72.5MiB/s][r=56.0k,w=18.6k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=11504: Sat Mar  2 15:53:12 2024
  read: IOPS=53.7k, BW=210MiB/s (220MB/s)(113GiB/548945msec)
   bw (  KiB/s): min=184032, max=238496, per=100.00%, avg=214950.29, stdev=8732.04, samples=1097
   iops        : min=46008, max=59624, avg=53737.59, stdev=2183.02, samples=1097
  write: IOPS=17.9k, BW=69.9MiB/s (73.3MB/s)(37.5GiB/548945msec); 0 zone resets
   bw (  KiB/s): min=61320, max=80000, per=100.00%, avg=71638.72, stdev=2931.50, samples=1097
   iops        : min=15330, max=20000, avg=17909.68, stdev=732.87, samples=1097
  cpu          : usr=7.29%, sys=86.91%, ctx=6151117, majf=0, minf=6
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=210MiB/s (220MB/s), 210MiB/s-210MiB/s (220MB/s-220MB/s), io=113GiB (121GB), run=548945-548945msec
  WRITE: bw=69.9MiB/s (73.3MB/s), 69.9MiB/s-69.9MiB/s (73.3MB/s-73.3MB/s), io=37.5GiB (40.3GB), run=548945-548945msec

Samsung PM863a 3.84TB TLC - SATA

Also very durable and affordable server SSDs

"Data center SSD drives will also work well." - absolutely

Solidigm P44 Pro 2TB (apparently same as SK Hynix Platinum P41)

fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.35
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=397MiB/s,w=132MiB/s][r=102k,w=33.8k IOPS][eta 00m:00s] 
test: (groupid=0, jobs=1): err= 0: pid=21095: Thu Mar 28 14:39:36 2024
  read: IOPS=99.0k, BW=387MiB/s (405MB/s)(113GiB/297984msec)
   bw (  KiB/s): min=369432, max=427240, per=100.00%, avg=396158.20, stdev=7717.86, samples=595
   iops        : min=92358, max=106810, avg=99039.48, stdev=1929.49, samples=595
  write: IOPS=33.0k, BW=129MiB/s (135MB/s)(37.5GiB/297984msec); 0 zone resets
   bw (  KiB/s): min=123927, max=142888, per=100.00%, avg=132031.68, stdev=2768.09, samples=595
   iops        : min=30981, max=35722, avg=33007.83, stdev=692.04, samples=595
  cpu          : usr=11.94%, sys=87.86%, ctx=5183, majf=0, minf=10
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=387MiB/s (405MB/s), 387MiB/s-387MiB/s (405MB/s-405MB/s), io=113GiB (121GB), run=297984-297984msec
  WRITE: bw=129MiB/s (135MB/s), 129MiB/s-129MiB/s (135MB/s-135MB/s), io=37.5GiB (40.3GB), run=297984-297984msec

Disk stats (read/write):
    dm-1: ios=29475801/9827719, merge=0/0, ticks=1779272/215384, in_queue=1994656, util=100.00%, aggrios=29492439/9833359, aggrmerge=0/0, aggrticks=1750608/206020, aggrin_queue=1956628, aggrutil=100.00%
    dm-0: ios=29492439/9833359, merge=0/0, ticks=1750608/206020, in_queue=1956628, util=100.00%, aggrios=29492409/9832110, aggrmerge=30/1266, aggrticks=1496885/97981, aggrin_queue=1594922, aggrutil=100.00%
  nvme0n1: ios=29492409/9832110, merge=30/1266, ticks=1496885/97981, in_queue=1594922, util=100.00%

Btw do we know if turning on disk encryption / LUKS under Ubuntu has a negative impact on IOPS?

Another Samsung SSD 990 Pro 4TB here, pretty much same results as @SnoepNFTs
https://gist.github.com/yorickdowne/f3a3e79a573bf35767cd002cc977b038?permalink_comment_id=4958391#gistcomment-4958391

It came preinstalled with the current latest firmware, 4B2QJXD7.

I might want to look at cooling too under actual staking load, got up to 61 degrees for fio (thermal pad goes to chassis).

With PCIe3 Nuc, booted from the drive

Intel Core i5-8259U, 1x8 GB ram.
Standard setup Ubuntu 22 server, LVM2 - Ext4 root partition.
Testing while booted from the same drive.

$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=528MiB/s,w=175MiB/s][r=135k,w=44.9k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=4836: Fri Mar 29 18:28:36 2024
  read: IOPS=136k, BW=530MiB/s (556MB/s)(113GiB/217450msec)
   bw (  KiB/s): min=530312, max=550488, per=100.00%, avg=542804.63, stdev=3349.45, samples=434
   iops        : min=132578, max=137622, avg=135701.18, stdev=837.40, samples=434
  write: IOPS=45.2k, BW=177MiB/s (185MB/s)(37.5GiB/217450msec); 0 zone resets
   bw (  KiB/s): min=175856, max=184608, per=100.00%, avg=180905.97, stdev=1344.54, samples=434
   iops        : min=43964, max=46152, avg=45226.49, stdev=336.13, samples=434
  cpu          : usr=25.22%, sys=64.23%, ctx=9698025, majf=0, minf=8
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=530MiB/s (556MB/s), 530MiB/s-530MiB/s (556MB/s-556MB/s), io=113GiB (121GB), run=217450-217450msec
  WRITE: bw=177MiB/s (185MB/s), 177MiB/s-177MiB/s (185MB/s-185MB/s), io=37.5GiB (40.3GB), run=217450-217450msec

Disk stats (read/write):
    dm-0: ios=29484828/9831160, merge=0/0, ticks=1113068/84136, in_queue=1197204, util=100.00%, aggrios=29492392/9837065, aggrmerge=0/347, aggrticks=1129394/104103, aggrin_queue=1233551, aggrutil=99.99%
  nvme0n1: ios=29492392/9837065, merge=0/347, ticks=1129394/104103, in_queue=1233551, util=99.99%

Above but with nvme_core.default_ps_max_latency_us=0 +3k/+1k

$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=541MiB/s,w=178MiB/s][r=138k,w=45.6k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1520: Fri Mar 29 18:41:21 2024
  read: IOPS=139k, BW=543MiB/s (569MB/s)(113GiB/212288msec)
   bw (  KiB/s): min=541560, max=564745, per=100.00%, avg=556696.58, stdev=4091.32, samples=424
   iops        : min=135390, max=141186, avg=139174.08, stdev=1022.81, samples=424
  write: IOPS=46.3k, BW=181MiB/s (190MB/s)(37.5GiB/212288msec); 0 zone resets
   bw (  KiB/s): min=180120, max=188672, per=100.00%, avg=185536.00, stdev=1475.02, samples=424
   iops        : min=45030, max=47168, avg=46383.90, stdev=368.75, samples=424
  cpu          : usr=25.22%, sys=66.61%, ctx=9314979, majf=1, minf=8
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=543MiB/s (569MB/s), 543MiB/s-543MiB/s (569MB/s-569MB/s), io=113GiB (121GB), run=212288-212288msec
  WRITE: bw=181MiB/s (190MB/s), 181MiB/s-181MiB/s (190MB/s-190MB/s), io=37.5GiB (40.3GB), run=212288-212288msec

Disk stats (read/write):
    dm-0: ios=29465825/9850565, merge=0/0, ticks=1110740/86064, in_queue=1196804, util=100.00%, aggrios=29492604/9913113, aggrmerge=70/769, aggrticks=1195797/274537, aggrin_queue=1470392, aggrutil=100.00%
  nvme0n1: ios=29492604/9913113, merge=70/769, ticks=1195797/274537, in_queue=1470392, util=100.00%

---

Update: Running from newer hardware. (+55k/+18k compared to before).
AMD Ryzen 5 7535HS, 32GB DDR5, PCIe4.
Booting from drive, same OS install, without kernel flags.

$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=753MiB/s,w=250MiB/s][r=193k,w=64.1k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=1757: Mon Apr  1 16:17:35 2024
  read: IOPS=191k, BW=744MiB/s (780MB/s)(113GiB/154814msec)
   bw (  KiB/s): min=709368, max=786112, per=100.00%, avg=762292.92, stdev=12028.39, samples=309
   iops        : min=177342, max=196528, avg=190573.26, stdev=3007.09, samples=309
  write: IOPS=63.5k, BW=248MiB/s (260MB/s)(37.5GiB/154814msec); 0 zone resets
   bw (  KiB/s): min=236904, max=262448, per=100.00%, avg=254056.41, stdev=4101.30, samples=309
   iops        : min=59226, max=65612, avg=63514.10, stdev=1025.31, samples=309
  cpu          : usr=11.90%, sys=75.15%, ctx=9206371, majf=1, minf=6
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=744MiB/s (780MB/s), 744MiB/s-744MiB/s (780MB/s-780MB/s), io=113GiB (121GB), run=154814-154814msec
  WRITE: bw=248MiB/s (260MB/s), 248MiB/s-248MiB/s (260MB/s-260MB/s), io=37.5GiB (40.3GB), run=154814-154814msec

Disk stats (read/write):
    dm-0: ios=29448145/9814651, merge=0/0, ticks=1062404/66404, in_queue=1128808, util=99.98%, aggrios=29492582/9829489, aggrmerge=0/54, aggrticks=1120690/77989, aggrin_queue=1198724, aggrutil=99.94%
  nvme0n1: ios=29492582/9829489, merge=0/54, ticks=1120690/77989, in_queue=1198724, util=99.94%

Above but using nvme_core.default_ps_max_latency_us=0. -2k/-0.4k

$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.28
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=727MiB/s,w=242MiB/s][r=186k,w=62.0k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=2419: Mon Apr  1 16:09:05 2024
  read: IOPS=189k, BW=739MiB/s (775MB/s)(113GiB/155805msec)
   bw (  KiB/s): min=720896, max=777048, per=100.00%, avg=757531.24, stdev=10137.71, samples=311
   iops        : min=180224, max=194262, avg=189382.84, stdev=2534.42, samples=311
  write: IOPS=63.1k, BW=246MiB/s (258MB/s)(37.5GiB/155805msec); 0 zone resets
   bw (  KiB/s): min=238728, max=260384, per=100.00%, avg=252469.48, stdev=3522.78, samples=311
   iops        : min=59682, max=65096, avg=63117.37, stdev=880.69, samples=311
  cpu          : usr=12.71%, sys=74.39%, ctx=9217428, majf=1, minf=8
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=739MiB/s (775MB/s), 739MiB/s-739MiB/s (775MB/s-775MB/s), io=113GiB (121GB), run=155805-155805msec
  WRITE: bw=246MiB/s (258MB/s), 246MiB/s-246MiB/s (258MB/s-258MB/s), io=37.5GiB (40.3GB), run=155805-155805msec

Disk stats (read/write):
    dm-0: ios=29470271/9822022, merge=0/0, ticks=1095304/69956, in_queue=1165260, util=99.99%, aggrios=29492568/9829469, aggrmerge=0/46, aggrticks=1119861/77973, aggrin_queue=1197872, aggrutil=99.95%
  nvme0n1: ios=29492568/9829469, merge=0/46, ticks=1119861/77973, in_queue=1197872, util=99.95%

Booting from external drive. +7k/+5k

Note this is also a different OS (Fedora desktop vs Ubuntu server).
No nvme_core.default_ps_max_latency_us=0 flag set afaik.

$ fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=64 --size=150G --readwrite=randrw --rwmixread=75
test: (g=0): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=64
fio-3.29
Starting 1 process
test: Laying out IO file (1 file / 153600MiB)
Jobs: 1 (f=1): [m(1)][100.0%][r=807MiB/s,w=268MiB/s][r=206k,w=68.5k IOPS][eta 00m:00s]
test: (groupid=0, jobs=1): err= 0: pid=5545: Mon Apr  1 11:20:05 2024
  read: IOPS=206k, BW=805MiB/s (844MB/s)(113GiB/143193msec)
   bw (  KiB/s): min=792296, max=844728, per=100.00%, avg=824277.65, stdev=8860.95, samples=286
   iops        : min=198074, max=211182, avg=206069.44, stdev=2215.25, samples=286
  write: IOPS=68.6k, BW=268MiB/s (281MB/s)(37.5GiB/143193msec); 0 zone resets
   bw (  KiB/s): min=263120, max=283256, per=100.00%, avg=274714.52, stdev=3272.99, samples=286
   iops        : min=65780, max=70814, avg=68678.63, stdev=818.26, samples=286
  cpu          : usr=9.60%, sys=61.20%, ctx=8753572, majf=0, minf=7
  IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
     issued rwts: total=29492326,9829274,0,0 short=0,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=64

Run status group 0 (all jobs):
   READ: bw=805MiB/s (844MB/s), 805MiB/s-805MiB/s (844MB/s-844MB/s), io=113GiB (121GB), run=143193-143193msec
  WRITE: bw=268MiB/s (281MB/s), 268MiB/s-268MiB/s (281MB/s-281MB/s), io=37.5GiB (40.3GB), run=143193-143193msec

Disk stats (read/write):
    dm-0: ios=29475893/9823846, merge=0/0, ticks=1120298/73131, in_queue=1193429, util=99.98%, aggrios=29492326/9829358, aggrmerge=0/28, aggrticks=1134762/75903, aggrin_queue=1210697, aggrutil=99.96%
  nvme0n1: ios=29492326/9829358, merge=0/28, ticks=1134762/75903, in_queue=1210697, util=99.96%

Standard setup Ubuntu 22 server, LVM2 - Ext4 root partition.

Thanks for sharing! Appreciate it! Also funny that the nvme_core.default_ps_max_latency_us=0 also slightly improved your performance. We can maybe add this modest improvement at the top of the .md? @yorickdowne

@SnoepNFTs likewise! :] interesting to compare.

For the kernel flag, I retested with a newer host (PCIe4 vs PCIe3, more ram, etc.). It's seemingly worse there.
So I'd probably call this within margin of error or situational.

If I understand right the flag is supposed to lock the power management of the drive to always be full power. Under a synthetic load I'd not expect this to really matter, the load should keep it pretty busy to be full power anyway. Maybe it matters in real workload once you're synced and data trickles in more randomly?

Btw, you mentioned changing systems didn't meaningfully change your iops but I got a pretty decent uplift.
Only thing I haven't tried yet is noatime, since this is the root filesystem need to do partition shuffling for that.

So here's one more PSA.

I tested Ext4, XFS, F2FS and for Ext4 poked at the noatime, even risky ones like data=writeback,barrier=0,nobh ( ⚠️ don't use those). I noticed one particular factor that caused major slowdowns...
Thermal throttling 😂

All of the settings and fs are within margin of error, not enough to care or worry about.
(The default relatime is more than good enough.)

But once I ran enough benchmarks that the SSD was hitting >70 degrees, I lost about 10k read and 3k write iops.
Which came back after letting it cool to about 40 degrees.

I for one am going to keep using Ext4 with relatime, not bothering with a "tuned data partition" of any kind.
But the SSD temps are definitely going to be on my grafana dash.

Should the other components of the setups be included in reports ?

The Kingston KC3000 perf is wonderful, but I'd love to know about the rest of the config. cc @kaloyan-raev.

Also, has there been any exploration regarding PCIe expansion cards such as the ASUS Hyper M.2 ?

The reasoning being that instead of creating a system with a single 4TB SSD (which usually uses 4 PCIe lanes if M.2), create one with four 1TB SSDs as well as a PCIe expansion card. This would allow for using 16 PCIe lanes, potentially improving IOPS and read/write speeds, for the same price or less.

The main thing this will accomplish is quadruple the failure rate and make it more difficult to change the size if ever desired, as CloneZilla doesn’t understand lvm or raid0.

Fast enough is fast enough. On a single drive Geth syncs in 6 hours and Nethermind can attest after 2h. Steady state both are well under 500ms for new blocks.