Connections dropped under load

[George Georgovassilis]

Many thanks for the pointer!

If I understand this correctly, there are some implications following 
session_linger:

1. High values require also large thread pools to make up for the 
lingering sessions ?
2. Low values are safer but may result in increased CPU usage ?
3. The effectiveness of session_linger depends on the network latency: 
if requests are piped in at a slow rate more sessions are locked up waiting?

If 3 is correct then session_linger sounds like a dangerous toy, because 
I can't really control the network latency.

Regards,
G.

On 06.01.2011 01:00, Poul-Henning Kamp wrote:
> In message<4D250296.2000601 at gmail.com>, George Georgovassilis writes:
>
>> 1. The benchmark replicates an expected real-life sequence of requests
>> and workload from a single IP (namely a corporate web-proxy), thus
>> labelling it "synthetic" does it no justice :-)
> Well, that depends on your proxy more than anything else, but I'll
> take your word for it.
>
>> 2. If you leave "session_linger" out of the configuration (so not
>> mentioning it at all) the benchmark still hangs. Whatever the default
>> value is, it doesn't work and I explicitly need to reduce it to 20.
> The default is 50msec.
>
> You can always get the full spiel, inclusive the default value
> from the CLI:
>
> param.show session_linger
> 200 1031
> session_linger             50 [ms]
>                             Default is 50
>                             How long time the workerthread lingers on the
>                             session to see if a new request appears right
>                             away.
>                             If sessions are reused, as much as half of all
>                             reuses happen within the first 100 msec of the
>                             previous request completing.
>                             Setting this too high results in worker threads
>                             not doing anything for their keep, setting it too
>                             low just means that more sessions take a detour
>                             around the waiter.
>
>                             NB: We do not know yet if it is a good idea to
>                             change this parameter, or if the default value is
>                             even sensible.  Caution is advised, and feedback
>                             is most welcome.
>
>