CPU usage is 100% when _close_sys() call is made

[wilson vivek]

Hi HP Support,

CPU spike is seen (90% for a 2 CPU HPUX ia64 11.23) when we make a close() socket call for a connection that is closed from the client end. Here is the stack frame for the thread that is hogging the CPU.

0: c00000000031f2f0 : _close_sys() + 0x30 (/usr/lib/hpux64/libc.so.1)
1: c00000000032f8c0 : close() + 0xa0 (/usr/lib/hpux64/libc.so.1)
2: c0000000484b3140 : _ZN6ACE_OS11closesocketEi() + 0x40 (/home/nparama1/IS_1118954721_benz_8.6.1.EE_64/8.6.1.EE/server/bin/libACE.sl)
3: c0000000484b3040 : _ZN8ACE_SOCK5closeEv() + 0x60 (/home/nparama1/IS_1118954721_benz_8.6.1.EE_64/8.6.1.EE/server/bin/libACE.sl)
We also found a posting in Google regarding the same behavior (http://www.unix.com/302086144-post1.html) but could not find any answers for this.
Can you please give pointers to resolve this.

Thank You
Wilson
Informatica Technical Support

[Dennis Handly]

>Hi HP Support,

This isn't an official support channel. You'll need to contact the Response Center.

>Can you please give pointers to resolve this.

Perhaps the kernel is flushing buffers or something?
Are you update on patches?

[wilson vivek]

Well, The patches aren't up-to-date

What we have is

BUNDLE11i B.11.23.0409.3 Required Patch Bundle for HP-UX 11i v2 (B.11.23), September 2004

Should i ask my admin to update to

BUNDLE11i (B.11.23.0409.3)

Thank You for your help
Wilson

[wilson vivek]

Also can you let me know if there is any specific patch you are looking for? Please let me know.

Thank You
Wilson

[Dennis Handly]

>Should I ask my admin to update to BUNDLE11i (B.11.23.0409.3)

I don't see how that would be better since it is the same. Why not try the latest, Dec '09 (11.23 Support Pack)?
http://www.itrc.hp.com/service/patch/releaseIndexPage.do

>can you let me know if there is any specific patch you are looking for?

A hang or loop on close?
Possibly: PHNE_36575:
QX:QXCR1000587628 CR:JAGag42702 SR:8606490242
Close operations on streams-based pipes appear to be hung in _csq_acquire() with the following stack trace and does not make forward progress.

[wilson vivek]

Hi,
This patch did not help in resolving the issue.
Here is the summary of the actual problem that has been reported.
We have a server running on HPUX and a client connection is made from a windows box.
Now when I remove the client machine from the network and the server code has a logic to wait for reconnection for a specified timeout (3 Minutes) and then tries to close the connection.
This is when we are seeing the CPU Spike and the stack shows the function calls.
Further, when I look into the netstat -a output, I could see that the connection from the client are marked as ESTABLISHED even after 7 Minutes from the disconnect. This is the period when we are seeing the CPU Spike.
Further once the connection is closed after 8 minutes (+ 3 Minutes for the application to detect a broken connection) , the CPU utilization is brought down.
Can you help with this?
Thank You
Wilson

[Dennis Handly]

>This patch did not help in resolving the issue.

You could try a patch assessment:
http://www.itrc.hp.com/service/patch/assessSystemsPage.do

[wilson vivek]

Hi,

Do you think we can tweak any of the tcp parameters so that the OS Detects the connection failure soon?

Thank You
Wilson

[wilson vivek]

Hi,

Attached is the patch assesment.

-Wilson

[Dennis Handly]

>Do you think we can tweak any of the tcp parameters so that the OS Detects the connection failure soon?

Rick, a network expert would know more.

>Attached is the patch assessment.

You didn't do: step 3. run the assessment and view patch recommendations ?

It basically recommends QPK1123(B.11.23.0912.082) and 6 others.

You have an unofficial patch UNOF_JAGAF30549_1 that seems fixed in PHKL_37725.