I'm doing some performance testing on our server and what I'm seeing
is a little disappointing.
We're running JDK1.2.2-RC4 (native threads) on several different
machines. I see almost identical behavior running on both of the
following machines:
quad 200MHz PII, 2GB memory, glibc 2.1.3, kernel 2.3.51
single 500MHz pIII, 256MB, glibc 2.1.2, kernel 2.2.12 (stock RedHat 6.1)
FYI, this same codebase runs on NT, Solaris and HP-UX, and it scales
quite well.
Our application server starts rejecting TCP connections (actually,
the OS does, our app never sees these rejected requests) at about 20
simultaneous requests. The load average on the system runs a little
high (5-10) when requests start getting bounced, but not overly
concerning. Connections start getting bounced within about 10
seconds of starting the load test.
The test page that we are hitting is fairly simple (it just displays
the first few rows of a DB table). Making the task more complex
(getting 30,000 rows) doesn't significantly alter (help or hurt) the
number of connections we can do.
Is there a limit on the number of TCP connections that the VM can
handle? Is there an kernel tuning paramater that I don't know about?
Our server starts about 250 client threads in a pool to handle the
requests, and (as expected) most are idle when looking at a thread
dump. Changing the thread pool size to 40 makes no difference on the
behavior.
I'm thinking that the kernel TCP connection queues are filling up
and further requsts are getting dropped, but at 20?
Any ideas on how to debug this would be greatly appreciated!
-John
----------------------------------------------------------------
John Rousseau [EMAIL PROTECTED]
SilverStream Software
1 Burlington Woods Phone: +1 781 238 5564
Burlington, MA 01803 Fax: +1 781 238 5499
----------------------------------------------------------------
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
