ASR detected by system ROM during stress test.

[kenken_3]

Hi,

I have a DL585G1 linux box as a web servers. Currently we conduct a stress test. It always rebooted when the concurrent connection reachs 100. We have 4 CPUs. Before the system reboot, the first CPU reaches about 63% utilization, but the other 3 cpus are about 20% utilization. Major processes are java, tomcat and apache.

The system doesn't capture any useful inforamtion except "ASR detected by system ROM" after system reboot.

Anyone has encountered this before? please advise.

[Stuart Browne]

ASR's are caused when the system can't contact the OS via an NMI.

The 'hpasmd' which is running is what resets the ASR counter.

Using the 'hplog' tool, you can alter the ASR period, or even disable it.

That being said, only 100 concurrent connections? It should be getting ASR timeout's with so little load, unless the tomcat/apache services aren't really well configured.

Make sure you're running the most up to date packages available, and not the old distributed packages (which are usually fairly buggy).

Possibly change which JRE you're using, if you can do that. Most JRE's run in single thread unless you tweak it fairly hard to use multiple threads. From your description, your JRE is only running a single thread (thus why one CPU is getting most of the load, and the others are sleeping).

Now, all this being said, you've only talked about what sort of CPU utilization you're witnessing. A busy system is more than CPU usage. It's also tied to the IO subsystems, what sort of runqueues you're getting, memory and swap usage etc. etc. etc.

Use tools like 'vmstat' and 'sar' to help monitor these things is helpful. Set up 'snmp' and MRTG/Cacti to graph the results to make sense of them.

Keep poking it.. ;)

[Bharath_Pingali]

Hello KenKen,

As a first step to debug, I would disable ASR thru BISO.

This should avoid the system rebooting and give a chance to debug.

-Good Luck
Bharath

[kenken_3]

I have tried to disable it, but the system just hung there and I still can't get any valueble information.