[
https://issues.apache.org/jira/browse/HTTPCORE-368?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13848034#comment-13848034
]
offbynull edited comment on HTTPCORE-368 at 12/13/13 11:00 PM:
---------------------------------------------------------------
Hi Oleg,
We based this information of off heap analysis that we did on our tool. The
buffers are taking up the largest chunk of available memory. Each buffer took
around 16k. There's 4 buffers being held. We're attempting to sustain somewhere
around 35k to 55k connections.
So for 35k connections, that's 16k * 4buffers * 35,000connections. That's
around 2+gigs of heap.
Before the patch, we got OutOfMemoryExceptions around 35-37k connections. After
the patch, we were able to sustain around 55k connections (we haven't tried
pushing this limit any higher).
In any event, does it make sense for buffers to exist when nothing is making
use of them? I understand that this may not be the best way to get around this
problem. I've heard from my colleague that "Netty makes use of an SslBufferPool
to reduce memory consumption during SSL communication" (his words). A pool of
buffers may be a good workaround to memory constraints as well, although this
may introduce some extra complexity around threading.
was (Author: offbynull):
Hi Oleg,
We based this information of off heap analysis that we did on our tool. Each
buffer took around 16k. There's 4 buffers being held. We're attempting to
sustain somewhere around 35k to 55k connections.
So for 35k connections, that's 16k * 4buffers * 35,000connections. That's
around 2+gigs of heap.
Before the patch, we got OutOfMemoryExceptions around 35-37k connections. After
the patch, we were able to sustain around 55k connections (we haven't tried
pushing this limit any higher).
In any event, does it make sense for buffers to exist when nothing is making
use of them? I understand that this may not be the best way to get around this
problem. I've heard from my colleague that "Netty makes use of an SslBufferPool
to reduce memory consumption during SSL communication" (his words). A pool of
buffers may be a good workaround to memory constraints as well, although this
may introduce some extra complexity around threading.
> Smarter use of buffers in SSLIOSession
> --------------------------------------
>
> Key: HTTPCORE-368
> URL: https://issues.apache.org/jira/browse/HTTPCORE-368
> Project: HttpComponents HttpCore
> Issue Type: Improvement
> Components: HttpCore NIO
> Affects Versions: 4.3-beta2
> Reporter: offbynull
> Labels: patch
> Attachments: SSLIOSession.patch
>
>
> We're using the async client to load test our comet infrastructure. We're
> frequently running in to OutOfMemoryErrors due to the ByteBuffers allocated
> in SSLIOSession.
> The following is a quick-and-dirty patch to SSLIOSession that removes the
> buffers when they aren't needed. Could we get this (or something similar to
> this) added in to httpcore-nio?
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
