[ejabberd] some ejabberd memory benchmarks

[Jack Moffitt]

>>> We are noticing similar things, are these tls connections? I ask because
>>> searching for this I found the following ticket.
>>>
>>> https://support.process-one.net/browse/EJAB-647
>>
>> Yep, they are. I'll do another set of tests disabling it and see what happens.
>
> Not the same situation as in the ticket, but disabling tls dropped
> memory usage of almost 30%

So for us, memory usage is around 1GB with only 140 people connected.
We're using odbc backends for everythingi but mod_last.  None of the
mnesia tables is very large (largest one is 7MB and is offline_msg).

Inspecting the running processes and memory shows just under 600
processes, the largest one is between 30 and 128MB in size usually.
That is often dets:open_file_loop2.  So there seem to be many fairly
large processes which eventually add up to a lot of memory usage.  At
peak load today this was regularly causing the server (2GB of ram) to
run out of memory, and crashed ejabberd.

Is this a known problem, or is there a leak somewhere?  For now we
moved to a machine with 8GB of ram.

jack.