Todd Lipcon has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/16189 )

Change subject: client: allocate InFlightOp from a Batcher Arena
......................................................................


Patch Set 1:

(1 comment)

http://gerrit.cloudera.org:8080/#/c/16189/1/src/kudu/client/batcher.cc
File src/kudu/client/batcher.cc:

http://gerrit.cloudera.org:8080/#/c/16189/1/src/kudu/client/batcher.cc@379
PS1, Line 379:     int ifo_prefetch = i + kPrefetchDistance * 3;
             :     int op_prefetch = i + kPrefetchDistance * 2;
             :     int row_prefetch = i + kPrefetchDistance;
             :     if (ifo_prefetch >= 0 && ifo_prefetch < size) {
             :       __builtin_prefetch(ops_[ifo_prefetch], 0, 
PREFETCH_HINT_T0);
             :     }
             :     if (op_prefetch >= 0 && op_prefetch < size) {
             :       const auto* op = ops_[op_prefetch]->write_op.get();
             :       if (op) {
             :         __builtin_prefetch(&op->row().isset_bitmap_, 0, 
PREFETCH_HINT_T0);
             :       }
             :     }
             :     if (row_prefetch >= 0 && row_prefetch < size) {
             :       const auto* op = ops_[row_prefetch]->write_op.get();
             :       if (op) {
             :         __builtin_prefetch(op->row().isset_bitmap_, 0, 
PREFETCH_HINT_T0);
             :       }
> Just making sure I understand this bit: while ops_ are all allocated in the
yea, and even within the same arena, they're allocated in the order the ops are 
submitted, but in the case that the ops are strewn about more than one tablet 
due to partitioning, the ops associated with a single WriteRpc will not 
necessarily be entirely contiguous/sequential in this loop, so the prefetching 
is probably still effective/useful.



--
To view, visit http://gerrit.cloudera.org:8080/16189
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I446a8d21253b7a274872bff6d3e76705ac95d0d5
Gerrit-Change-Number: 16189
Gerrit-PatchSet: 1
Gerrit-Owner: Todd Lipcon <[email protected]>
Gerrit-Reviewer: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Andrew Wong <[email protected]>
Gerrit-Reviewer: Andrew Wong <[email protected]>
Gerrit-Reviewer: Bankim Bhavsar <[email protected]>
Gerrit-Reviewer: Kudu Jenkins (120)
Gerrit-Reviewer: Todd Lipcon <[email protected]>
Gerrit-Comment-Date: Tue, 14 Jul 2020 03:10:09 +0000
Gerrit-HasComments: Yes

Reply via email to