Created
April 1, 2011 23:48
-
-
Save PharkMillups/899066 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
<TB123> Hi, I'm trying to do some benchmarking on Riak | |
and can't load more than 0.6Gbits write (approx 3MBs messages), | |
bitcask storage. Is it normal, or should I concentrate on tunnig some parameters? | |
13:17 <TB123> there is only 4GB RAM, maybe that could be problem | |
13:19 <TB123> Gbit/s | |
13:23 <seancribbs> each message is 3MB? | |
13:27 <TB123> @sean: Yes. | |
13:28 <seancribbs> so, some questions. 1) how many nodes are you | |
running? 2) how are you running the benchmark | |
13:28 <seancribbs> 3) are you using the default n_val | |
13:30 <TB123> 1) currently on 1 node, to tune the configuration | |
and than will move to 8 nodes to test scaling b) one java PBC | |
client with 10 threads connecting over 1Gbit net to the other. c) yes. default | |
13:30 <TB123> add b) have tried to load from two different servers on one time, but with same result | |
13:32 <TB123> sorry, from two different client | |
14:02 <seancribbs> TB123: sorry for the delay. I think you will | |
find performance less than expected until you reach 4+ nodes | |
14:02 <seancribbs> we recommend 5 nodes to start, absolute minimum of 3 | |
14:08 <TB123> @seanscribbs: OK. I'll do the same test on 6 | |
nodes. My reason for testing on one node was to understand, | |
what is the throughput limit and to tune performance on one place (node). | |
14:10 <TB123> @seanscribbs: I assume, this 0.6Gbp/s limit is | |
caused by internal riak overhead, memory locks, etc. | |
14:22 <seancribbs> TB123: not locks per se, but yes, single-systems can become bottlenecks | |
14:26 <TB123> @seancribbs: Ok, understad. Thanks. | |
14:26 <seancribbs> when you're at multiple nodes, spread the request | |
load to all nodes too, perhaps with a LB | |
14:28 <TB123> for my benchmarking purposes, i'll connect same | |
number of threads to every node. I don't currently have any loadbalancer prepaired. | |
14:31 <seancribbs> TB123: sounds good |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment