Cache size optimizations #1981
Cache size optimizations #1981
Conversation
|
ACK |
|
Automatic sanity-testing: PASSED, see http://jenkins.bluematt.me/pull-tester/1c83b0a3771bc601fdc75588f2cd45318b19c526 for binaries and test log. |
|
ACK |
|
I'm interested, what benefit does this induce performance wise? Any numbers or some more details? |
|
Before, the LevelDB write buffer was always 4 MiB, the LevelDB cache for both databases (one very small, one somewhat larger), and the coin cache was flushed everytime 5000 transactions changed. This commit makes the -dbcache value better respected, by splitting it over all types of caches, using a meaningful distribution (I tried a few variations). The result is far fewer flushes during IBD, lower memory usage (as it's 25 MiB total now, instead of 2x 25 MiB in addition to the coin cache and write buffer) and the available cache size is used more evenly. Performance-wise: I got a 1.4x speedup for a -reindex to block 193000. |
Cache size optimizations
be3ae1b [Cleanup][Sapling] Remove uint252 class (random-zebra) 1353507 [Cleanup] Remove Sprout pseudo-random functions (random-zebra) 1c23279 [Cleanup] Remove old NoteEncryption/NoteDecryption classes (random-zebra) 52d2625 [Cleanup] Remove Sprout address/key classes (random-zebra) 09875f0 [Cleanup] Remove SproutNote class (random-zebra) 8747949 [Cleanup][RPC] Remove unused fields in validateaddress output (random-zebra) Pull request description: Remove all sprout-related code. ACKs for top commit: furszy: code reviewed ACK be3ae1b . Fuzzbawls: Code/Tests ACK be3ae1b Tree-SHA512: 8354adb14372126495a069c0c4aa9e0ee4c400ebf8cdff4e120a67c0e3ea1a190d1617645bc03e02e8370ab644c14769232ae8cef57d6230adc0095ac70641dd
Tweaks for LevelDB's caches.