Too many disk writes during blockchain sync

[nimnul]

Has this been discussed before? Bitcoin client seems to perform way too many disk writes under Windows - Task Manager reports 4 millions of write calls to sync 3k blocks.

Can't we buffer writes to bring them down to dozens per second from hundreds per second? My hard drive seeks like crazy during sync.

[1715282234]

1715282234

[1715282234]

1715282234

[Pieter Wuille]

Please try v0.8.0rc1.

[nimnul]

It used 72M writes to reindex Still reindexing

[grue]

use 0.8 rc with a large cache size (-dbcache=1024)

[nimnul]

It takes 20M+ writes to start with -dbcache=1024. Indexing is not over yet.

[gmaxwell]

It takes 20M+ writes to start with -dbcache=1024. Indexing is not over yet.

Then it sounds like 'write calls' doesn't mean what we think it means here.... or leveldb is doing something crazy on win32. In strace output it's pretty clear that it's being batched as expected under Linux.

[nimnul]

It has the same meaning as strace write syscall count.

How many writes I'm supposed to get? If the issue is supposed to be fixed I can look closer at it.

Now it finished reindexing and syncing remaining 2k blocks or so, used "only" 30k writes to start and then about 10k writes per block.