On 07/23/2012 09:13 PM, Mladen Adamovic wrote:
> On Mon, Jul 23, 2012 at 7:30 PM, Oleksiy Stashok
> <oleksiy.stashok_at_oracle.com <mailto:oleksiy.stashok_at_oracle.com>> wrote:
>
>> the changes from default config I have increased
>> http-thread-count to 300 (with default number 8 problems with
>> accepting connection were immediately evident, although not too
>> many open files problem was evident in the log) and also added
>> MySQL connection polling to avoid problem with many connections
>> in TIME_WAIT state, see http://lists.mysql.com/java/6586
> That's strange, if average processing time is 3ms, 8 threads
> should be more than enough... if i'm not missing anything obvious
> here.
>
>
> I think that perhaps, the average processing time doesn't take into
> account the time it waits for connections or stalled time. Think about
> keep alive connections.
> Normally, I could see 80 active keep alive connections.
> When keep alive connection pulls a request, one RTT is needed until
> next command is received over the same connection, so between requests
> it can be 300ms waiting time for RTT.
> And if all keep alive connections are in the queue, the server might
> wait for RTT in all 8 threads processing keep alive connections.
> At the same time queue is getting overloaded with connections... If
> I'm not missing anything different here?
Once HTTP request is processed (if HTTP pipelining is not used) - we
release a worker thread and it's ready to process another HTTP request
on different connection.
We don't keep worker thread bound to a specific connection, even if it's
in keep-alive mode.
> It might be related to the socket linger timeout, pls. check
> what's the default linger value for your env.
> http://stackoverflow.com/questions/9563496/how-do-i-find-out-the-jre-default-linger-on-close-time-out-value
>
>
> It shows: Socket.getLinger() = -1
Ok, looks like it's not linger related.
When you disable keep-alive connections, does it change anything w.r.t.
a number of CLOSE_WAIT connections?
Thanks.
WBR,
Alexey.
>
> Hm, I've seen one problem with the Virtuozzo server in the log:
> quotaugidlimit
> Number of user/group IDs allowed for the Container internal disk
> quota. If set to 0, UID/GID quota will not be enabled.
>
> This is set to limit 2000, but I don't understand what it has with
> files, it shall be number of UID/GIDs, and at the moment is has been
> steady at 43 at VPS.
> Perhaps these are problem with Virtuozzo setup but it has to be proven
> somehow.
>
>
>
> --
> Mladen Adamovic
> Numbeo
> Drziceva 9, 11120 Belgrade-Zvezdara, Serbia
> (Business Registration Number 62612240)
> Tel. +381-66-058-595
> email: mladen.adamovic_at_gmail.com <mailto:mladen.adamovic_at_gmail.com>
> web: http://www.numbeo.com