[ejabberd] Performance Issues

Bill Omer bill.omer at gmail.com
Wed Dec 12 20:57:20 MSK 2007


I am using the latest SVN of ejabberd and jabsimul to stress test my
test cluster.  I am using the watchdog feature, and it is constantly
sending me messages like this:

(12:42:31 PM) ejabberddev01.ejabberddev: (ejabberd at ejabberddev03) The
process <0.443.0> is consuming too much memory:
[{heap_block_size,1346269},{mbuf_size,0},{stack_size,35},{heap_size,357336}].
[{current_function,{mnesia_tm,async_send_dirty,6}},
 {initial_call,{proc_lib,init_p,5}},
 {message_queue_len,1173},
 {links,[<0.227.0>]},
 {dictionary,[{random_seed,{17008,18662,20000}},
              {'$ancestors',[ejabberd_sup,<0.39.0>]},
              {'$initial_call',{gen,init_it,
                                    [gen_server,
                                     <0.227.0>,
                                     <0.227.0>,
                                     {local,
                                         'ejabberd_mod_pubsub_im2.ejabberddev'},
                                     mod_pubsub,
                                     ["im2.ejabberddev",
                                      [{access_createnode,pubsub_createnode},
                                       {plugins,["default","pep"]}]],
                                     []]}}]},
 {heap_size,1346269},
 {stack_size,35}]


My cluster consists of three servers.  The master server is not listed
in my DNS round robin, so all connections only go to the two slave
nodes.  The slave nodes' hardware is:  Dual Core AMD 1.8GHz with 2
gigs of ram.

I'm using jabsimul, with my xml config file attached.

Current memory usage:

Node1:
root at ejabberddev02:~# free -m
             total       used       free     shared    buffers     cached
Mem:          2012       1115        896          0        114         71
-/+ buffers/cache:        929       1082
Swap:           47          0         47


Node2:
root at ejabberddev03:~# free -m
             total       used       free     shared    buffers     cached
Mem:          2012       1120        891          0        132         73
-/+ buffers/cache:        914       1098
Swap:           47          0         47


CPU usage stays around <5% but does spike up to 50% at times.

Load averages:

Node1:
load average: 0.28, 0.29, 0.28

Node2:
load average: 0.37, 0.31, 0.27



I have 5000 active connections, sent almost 1.2 million messages,
running for 02:23:45.   Here's a screen grab of jabsimul



02:23.54
Conn stat:  conns:  total: 9194  estabilished: 5000
            kills:  total: 4194  unexpected:   2688
Messages:   tot.sent:     1198798        tot.rcvd:   1199693
            rcvd.offline: 0      rcvd.admin: 0
            rcvd.normal:  1199693        fwd: 599414     avg.time:   822 [ms]
            diff check:   -895   stability:  253
Roster:     tot.adds:     47288  avg.time:   3698 [ms]
            tot.dels:     5262   avg.time:   3849 [ms]   glob_rost: 602
Presences:  tot.sent:     605956         tot.rcvd:   3562872
Packets:    created:      1987066            sent:   1984618
           canceled:     2379   in queues:   69


The unexpected kills happened early, when I stopped a node and then
restarted it.

Basically, I'm wondering if the pubsub warnings from watchdog are
something I should be concerned about or not.





-- 
Regards,
Bill Omer
-------------- next part --------------
A non-text attachment was scrubbed...
Name: jab.xml
Type: text/xml
Size: 1912 bytes
Desc: not available
Url : http://lists.jabber.ru/pipermail/ejabberd/attachments/20071212/f9807dc9/attachment.bin 


More information about the ejabberd mailing list