-
Notifications
You must be signed in to change notification settings - Fork 28
[Keep open] Benchmark log #31
Comments
Commit b1e626f, go 1.11, after running
|
After optimising benchmark code, go 1.11:
|
After migrating to TxnDatastore, go 1.11, with transactions in closures:
|
Procedural style, go 1.11 -- no practical difference. Will make use of the functional style.
|
commit 686aaca
|
Big improvement in Commit raulk@03ee26f
|
Changed the
|
Commit raulk@903aa12 gives us close to a 30-40% reduction in cache-bound ClearAddrs.
|
benchcmp between baseline (prior to refactorings) and #34.
|
@raulk how do these numbers look with leveldb? |
@whyrusleeping very good question, my friend! Also, we migrated to From the point of view of the peerstore, I don't think we need transactional guarantees anyway. @bigs do you remember the argument for requiring a transactional DB over a batch-capable one? |
Ok, I see go-ds-leveldb uses go-leveldb and not the native leveldb. go-leveldb does support transactions, so I'll open a PR in go-ds-leveldb to make it implement |
@whyrusleeping Benchmarks for badger vs. leveldb vs. in-memory. Numbers are for the version before implementing datastore-native TTL (we'd need a shim for leveldb), but they provide a good baseline. For badger and leveldb, the numbers with and without the ARC cache are shown.
|
Badger seems to come out several multiples on top of the leveldb Go implementation. Testing against RocksDB (cgo) could be interesting. Also, the peerstore could benefit from an indexed DB like sqlite. |
good to know! i think rocksdb could be nice for the future, but i don’t think it needs to be a priority. i do however think a sqlite example could be worthwhile, though if all we really need from it is indexing, perhaps we could implement this in another badger db?
…On Oct 3, 2018, 2:03 PM -0400, Raúl Kripalani ***@***.***>, wrote:
Badger seems to come out several multiples on top of the leveldb Go implementation. Testing against RocksDB (cgo) could be interesting. Also, the peerstore could benefit from an indexed DB like sqlite.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub, or mute the thread.
|
@raulk would be a good idea to add |
No longer necessary to keep this trail since datastore-backed peerstore is now merged and performant. |
This issue tracks benchmark results as we continue tuning the peerstore. It will remain open until such efforts subside.
Please post the commit hash you're testing against and your CPU information.
The text was updated successfully, but these errors were encountered: