Any help for a relative noob?
I've set up a drupal server that uses nginx + php-fpm and boost. Boost and its rewrite rules are working correctly for anonymous users, as evidenced by the embedded boost comments in the pages I'm seeing.
Now I'm doing load tests, using jmeter with a simple test plan that has two thread groups, one ('Posters') that does a once only login, then creates a comment against a node, and another ('Readers') that hits the node. I am creating ten Poster threads, and 50 Reader threads, ramping up over 1 second, running them in parallel in a single test plan from a headless server on the same subnet.
The problem is that I'm seeing a very large number of errors, mostly among the Posters, sometimes as high as 100%. When I look at the results, many of the poster requests fail with a 'Peer closed connection' error. I've compiled nginx with-debug, and the nginx_error.log file does have some messages that seem to be correlated, but assign the blame to the client ('client XXXX.XXX closed keepalive connection (104: Connection reset by peer)'), and there aren't enough of them to account for the number of errors I'm seeing from the client.
My max_connections is fairly high - 1024 threads/process, and 6 processes. Further, I can't imagine I'm overwhelming the server, since I'm using boost for the readers, so not even hitting php for much of it, and we're only talking 60 concurrent connections on a mosso cloud server. Also, while I'm not sure it's even relevant, my keepalive_timeout is set at 60.
One more thing: I've temporarily disabled the form submit code in form.inc so I don't have to manage hidden form fields in my tests.
Any thoughts on how to approach this? I'm running out of ideas, and part of me is starting to wonder whether I'm chasing phantoms; I've yet to see a failure in the browser, where I've tried manually entering comments while the load test was running. I would very much like to feel confident that I'm pounding my server somewhere near as hard as my users soon will be, and that it's holding up well.

Comments
limit_conn
Do you have
limit_connset in your Nginx config?"Connection reset by peer" is a notice, nothing wrong, see: http://osdir.com/ml/web.nginx.english/2007-01/msg00041.html
Did you tried
ab(Apache Benchmark) test ?You can simulate logged in user with
ab -C DRUPAL_UID=1.I don't see any reason for that amount of errors here. Well, I also don't see your full Nginx config :) so it is pure guessing.
HTH
~Grace
-- Turnkey Drupal Hosting on Steroids -- http://twitter.com/omega8cc
Thanks for responding to this, Grace
I've put that project on hold for the time being, and now that I see things like Mercury appearing, I may not go back to it. As I was building the server, I kept asking myself why someone hasn't created a standard cloud image for drupal. Seems like a great community project.
I agree
Josh inspired me to finally start with something like that, but not limited to Amazon cloud.
Good questions are better than good answers :)
~Grace
-- Turnkey Drupal Hosting on Steroids -- http://twitter.com/omega8cc