Reducing bitcoind memory usage

reducing-bitcoind-memory-usage.md

There are a few parameters that can be dialed down to reduce memory usage of bitcoind. This can be useful on embedded systems or small VPSes.

In-memory caches

The size of some in-memory caches can be reduced. As caches trade off memory usage for performance, usually reducing these have a negative effect on performance.

• -dbcache=<n> - the UTXO database cache size, this defaults to 450 (since 0.14.1). The unit is MiB (1024)
  • Before 0.11.0 the computation here was off, so a factor of about five times more memory was used than specified.
  • The minimum value for -dbcache is 4.
  • A lower dbcache make initial sync time much longer. After the initial sync, the effect is less pronounced for most use-cases, unless fast validation of blocks is important such as for mining.

Memory pool

• To protect against mempool flooding set minrelaytxfee=0.00005 and limitfreerelay=5. See the 0.11 release notes at https://github.com/bitcoin/bitcoin/blob/v0.11.0/doc/release-notes.md#transaction-flooding . This is no longer necessary for 0.12 and above.

• In Bitcoin Core 0.12 and later there is a memory pool limiter which can be configured with -maxmempool=<n>, where <n> is the size in MB (1000), 300 by default.

  • The minimum value for -maxmempool is 5.
  • A lower maximum mempool size means that transactions will be evicted sooner. This will affect any uses of bitcoind that process unconfirmed transactions.

• To completely disable mempool functionality there is the option -blocksonly. This will make the client opt out of receiving (and thus relaying) transactions completely, except as part of blocks.

  • Do not use this when using the client to broadcast transactions as any transaction sent will stick out like a sore thumb, affecting privacy. When used with the wallet it should be combined with -walletbroadcast=0 and -spendzeroconfchange=0. Another mechanism for broadcasting outgoing transactions (if any) should be used.

• Since 0.14.0, unused memory allocated to the mempool (default: 300MB) is shared with the utxo cache, so when trying to reduce memory usage you should limit the mempool, with the -maxmempool command line argument.

Number of peers

• -maxconnections=<n> - the maximum number of connections, this defaults to 125. Each active connection takes up some memory. Only significant if incoming connections are enabled, otherwise the number of connections will never be more than 8.

Thread configuration

For each thread a thread stack needs to be allocated. By default on Linux, threads take up 8MiB for the thread stack on a 64-bit system, and 4MiB in a 32-bit system.

• -par=<n> - the number of script verification threads, defaults to the number of cores in the system minus one.
• -rpcthreads=<n> - the number of threads used for processing RPC requests, defaults to 4.

Linux specific

By default, since glibc 2.10, the C library will create up to two heap arenas per core. This is known to cause excessive memory usage in some scenarios. To avoid this make a script that sets MALLOC_ARENA_MAX before starting bitcoind:

#!/bin/bash
export MALLOC_ARENA_MAX=1
bitcoind

The behavior was introduced to increase CPU locality of allocated memory and performance with concurrent allocation. So this setting could in theory reduce performance. However, in Bitcoin Core very little parallel allocation happens, so the impact is expected to be small or absent.

Please note, since 0.14.1 the UTXO database cache size defaults to 450 MiB, not 300 MiB.

This is really cool. Thanks. Do you have any information about how these parameters affect the system's requirement and the total daily bandwidth use?

@danielalexiuc thanks, updated

@Sword-Smith lowering dbcache will make validation slower (on average) as more disk I/O to the database is needed as queries miss the cache. It will not have any effect on bandwidth use.
Reducing the maximum number of connections will likely reduce bandwidth use and lower CPU use.