Difference in behavior when enabling max-threads parameter

[Jakov <ja...@est.tech>]

Hi Apache Geode community,

Could you please help us understand Apache Geode /max-thread /parameter 
impact on traffic latency?

We are experiencing spikes in traffic latency when we use default value 
of/max-threads == 0/ but not when we set it to some value (in our case 
/max-threads = 64/). In both cases we limit client/server connections to 
/max-connections == 4000/. We have about a 100 clients each generating 
about 10k requests. And a cluster with 20 servers and 3 locators.

Does anybody knows how /max-threads/ affect Apache Geode that could 
cause major difference in latency for above cases?

Please don't hesitate to ask if you need any additional information.

BRs,

Jakov

Re: Difference in behavior when enabling max-threads parameter

[Dan Smith <ds...@pivotal.io>]

Hi Jakov,

I'm assuming you are talking about the max-threads setting on the cache
server. When you set max-threads, it is limiting the number of threads used
for processing requests from clients. Normally, the number of threads would
grow up to max-connections, or 4000 threads in your case. If you happen to
have conserve-sockets set to false, those 4000 threads will also trigger
the creation of additional peer-to-peer sockets and threads.

So, I would guess that at some point you are ending up with too many
threads for your servers to handle. Context switching between threads
definitely adds overhead, and if some threads don't get scheduled for a
while that could definitely increase your latency.

-Dan

On Fri, Mar 13, 2020 at 7:10 AM Jakov <ja...@est.tech> wrote:

> Hi Apache Geode community,
>
> Could you please help us understand Apache Geode *max-thread *parameter
> impact on traffic latency?
>
> We are experiencing spikes in traffic latency when we use default value of*
> max-threads == 0* but not when we set it to some value (in our case *max-threads
> = 64*). In both cases we limit client/server connections to *max-connections
> == 4000*. We have about a 100 clients each generating about 10k requests.
> And a cluster with 20 servers and 3 locators.
>
> Does anybody knows how *max-threads* affect Apache Geode that could cause
> major difference in latency for above cases?
>
> Please don't hesitate to ask if you need any additional information.
> BRs, Jakov
>
>
>