[
https://issues.apache.org/jira/browse/ARTEMIS-3045?focusedWorklogId=544158&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-544158
]
ASF GitHub Bot logged work on ARTEMIS-3045:
-------------------------------------------
Author: ASF GitHub Bot
Created on: 29/Jan/21 09:32
Start Date: 29/Jan/21 09:32
Worklog Time Spent: 10m
Work Description: franz1981 edited a comment on pull request #3392:
URL: https://github.com/apache/activemq-artemis/pull/3392#issuecomment-769689978
@michaelandrepearce There is any chance that this PR could be tried on a
real hardware?
I'm having some problem to get it ATM.
That should improve replication performance (especially with small sized
messages ie < MTU size): this PR try to batch records appends on the live
broker before sending it to the backup broker, maximizing the usage of the
outbound buffer provided by Netty.
That should help submitting the send requests saving potential signaling to
wake-up the event loop on each submitted packet.
In turn, this could help to better use the network bandwidth, saving from
sending near empty TCP packets.
Despite batching seems to trade-off latency vs troughput perf, here it
shouldn't happen, because:
- we send a batch of packets from the same thread, improving cache misses
while reading the batch of packets
- reducing the amount of syscalls per packet means amortizing the latency
cost per-packet, given that syscalls vectored writes (that Netty should use to
send the batches AFAIK) tends to have near fixed costs
My only concern is that impl-wise it could be implemented much better, as
mentioned on
https://github.com/apache/activemq-artemis/pull/3392#issuecomment-757663864 and
that batching could have some upper limit in the amount of bytes other then
Netty buffering that could be maybe too big.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
Issue Time Tracking
-------------------
Worklog Id: (was: 544158)
Time Spent: 3.5h (was: 3h 20m)
> ReplicationManager can batch sent replicated packets
> ----------------------------------------------------
>
> Key: ARTEMIS-3045
> URL: https://issues.apache.org/jira/browse/ARTEMIS-3045
> Project: ActiveMQ Artemis
> Issue Type: Improvement
> Reporter: Francesco Nigro
> Assignee: Francesco Nigro
> Priority: Major
> Time Spent: 3.5h
> Remaining Estimate: 0h
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
