Tomasz Sterna wrote:
> From: Eric Liang
>> I guess the designment is to get a static element container,  but
>> what about if there are a great amount users registered while just a
>> few of them are active? Meanwhile, the empty elements list will come
>> to a bottleneck while getting an element not exists in the hash table.
>
> Seems correct.
> Did you do any benchmarking?
Not yet.
for me , this is an issue more about memory usage than efficiency ,
while a hash table 's efficiency is much concerned with the collision
(or the real data), IMHO.  However, a smaller bucket-list is good for that.

>
>> two patches are attached, while the second one is to print the stat
>> of xhash, just for debug the initial size( sm->users/ sm->sessions
>> for example).
>
> Cool. Have you tested it under live conditions (with real users)?
Still testing. yes, under live conditions.

Actually,  I am puzzled by the memory using of _router_ which increasing
vastly these days. It looks like that the limit of _sm's_ throughput led
to a large size of _jqueue_ in _router._ The _jqueue_ is ok after
checking, but when I came to the _xhash,_ it looks not that fine.


>
> Could you create a new Bug on Launchpad
> https://bugs.launchpad.net/jabberd2/+filebug and attach the patches to
> it?
>
>

No problem, I'll do that asap.

There are more jabberd-servers today, Jabberd2 is still a good choice.
Thanks for your hard work.

Eric

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to