Optimizing IdP's performance

[Chad La Joie]

I have not looked closely at the APR connectors, that has had issues
of its own in the past (and they pop up again from time to time) and
it's a lot of JNI which, to me, is just voodoo of the most evil sort.

The main problem with the pure-Java connectors, when I looked at them,
was the lack of a dispatch queue.  Initially the blocking connector
spun up a thread per-reques, which pretty much caused the whole server
to die under load.  The non-blocking, and more recently the blocking,
connectors have had a thread pool that are used to answer requests.
However, they don't have a dispatch queue, which means that once all
the threads in the pool are in use you start getting 500 errors.  So
it doesn't degrade very gracefully.

I'll note, I haven't looked at the code in the last year or so (since
deciding we'll be using Jetty for the IdP), so v7 or perhaps more
recent v6 releases, might have addressed this issue.

On Mon, Aug 1, 2011 at 11:47, Marvin Addison <marvin.addison at gmail.com> wrote:
>> Tomcat's connection handling is pretty poor in my areas
>> and has a couple of high-latency sections (due to global or
>> semi-global locks).
>
> I'm curious about this comment.  Are you saying this is generally true
> and applies to all connectors (Coyote, NIO, APR) in your experience,
> or only certain ones?
>
> M
> --
> To unsubscribe from this list send an email to users-unsubscribe at shibboleth.net
>



-- 
Chad La Joie
www.itumi.biz
trusted identities, delivered