Hi,

We are doing some simple stress testing running ca 1500 requests (identical) over
a 60 second period repeatedly. Most of the time we see everything running smoothly
but for a 1 or 2 second interval in most of the tests we see throughput dropping to zero.

vmstat typicall shows

r b swpd free buff cache si so bi bo in cs us sy id wa

0 0 17444 3624 20060 100968 0 0 0 68 107 369 6 0 94 0
0 0 17456 4244 19280 100188 0 0 0 12 103 348 5 2 93 0
2 0 17456 3872 19280 100188 0 0 0 0 102 222 50 0 50 0
0 0 17456 3748 19280 100188 0 0 0 24 106 34792 34 16 50 0
0 0 17456 3748 19280 100188 0 0 0 0 96 348 6 1 93 0
2 0 17456 3748 19288 100180 0 0 0 76 101 353 12 1 85 2
0 0 17456 3748 19288 100180 0 0 0 0 105 360 8 1 91 0

for the period around the throughput slowdown. User cpu usage goes up from around 6 % to
50 or so. Context switches jump ca 100-fold. There is no unusual swap activity.

Top shows java -jar server.jar being responsible for the cpu utilization.

After a second or so throughput increases again to the expected levels.

Has anyone any ideas what may be happening ?

Platform is linux 64 bit (SLES 9.3) on IBM zseries. Application is a very simple jsp reading a
couple of rows from a db2 table and displaying the result, Nothing else is running on the server
and no other users.

Regards,

John.

=====================================
John Enevoldson

Reply via email to