How to achieve (1000 tx 1000 reads each)/sec on single client

I was looking again at the problem and found keynote about StorageCache process class. Will it help dramatically with performance and avoid “3 storage servers” bottleneck?

Yes, storage cache will bring more read replicas to improve the read performance on the same key (shard). @negoyal may have some performance numbers?

There is an issue to manually split a key range, which can help assign a small hot shard to more than 3 storage servers, which help both read and write throughput for the shard. But it won’t help with a single key performance.

We don’t have performance numbers yet, as we still are working to make this feature production ready. But the idea is to improve the performance in presence of read hot shards using more read replicas as @mengxu pointed out.

2 Likes

@sfc-gh-ngoyal I am looking at whether FDB supports read replica mode to scale read path. Is that something FDB supports today
?

@rmanoj It’s not yet supported. The work was paused last year in favor of other projects. So it’s not officially supported yet unfortunately.

1 Like