You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The basic idea and a rough benchmark result is posted here etcd-io/etcd#12026
And I've also tested this in raft-rs on a 5 node TiKV cluster. The result shows almost 30% reducing on append_resp messages in a bulk inserting work load.
I'm not sure the thread model of the network transport layer in sofa-raft. If the append response messages are sent asynchronously back to leader batch by batch (just like etcd), maybe we can gain some benefit by merging AppendResponse because I found sofa-raft free the Inflight's in a FIFO style.
The text was updated successfully, but these errors were encountered:
The basic idea and a rough benchmark result is posted here etcd-io/etcd#12026
And I've also tested this in
raft-rs
on a 5 node TiKV cluster. The result shows almost 30% reducing onappend_resp
messages in a bulk inserting work load.I'm not sure the thread model of the network transport layer in sofa-raft. If the append response messages are sent asynchronously back to leader batch by batch (just like etcd), maybe we can gain some benefit by merging AppendResponse because I found sofa-raft free the Inflight's in a FIFO style.
The text was updated successfully, but these errors were encountered: