[
https://issues.apache.org/jira/browse/TS-3547?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14509100#comment-14509100
]
Susan Hinrichs commented on TS-3547:
------------------------------------
Fiddling with the nagle times as [~sudheerv] suggested might also help.
May ultimately need to not use SSL_write but do BIO stuff directly to get the
control we need.
>From chatting with [~jacksontj] last night, it wasn't clear that the separate
>writes were really causing the delay. It sounded like the delays only
>occurred if the box was in rotation (under some load). But I see that you
>guys have done some more experiments since then. If you could share a pcap
>file of a pathological case, that would be great.
Also [~davet] might have some insights.
> SSLNetVConnection: switch to a vectored write
> ---------------------------------------------
>
> Key: TS-3547
> URL: https://issues.apache.org/jira/browse/TS-3547
> Project: Traffic Server
> Issue Type: Bug
> Reporter: Thomas Jackson
>
> UnixNetVConnection does a vectored write which bunches blocks together until
> the outgoing buffer will fill up. This means we can better fill the packets,
> and should give us a bit of a performance boost to SSL writes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
