fix?

[Prashant Kumar_1]

Hi Guys what is the fix then ..for the buffer problem on hp ux 11 ?

[Ian Dennison_1]

what specific buffer problem? if it is the 400MB usable limit of buffer cache, then stripe the disks, and use the extra memory for application buffers.

Share and Enjoy, Ian Dennison

[Prashant Kumar_1]

Sorry I should've posted the full problem .Well I've seen numerous replies on "buffer space unavailable" on HP UX 11 . I've a java application which fails on accept() when the load balancer does the health check . The error reported is " Socket Exception .. no buffer space " . There must be a fix out there .. help ?

Reference :
http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x700aa22d6d27d5118fef0090279cd0f9,00.html

[Bill McNAMARA_1]

mmm, I'd suggest getting faster CPUs to help clear your buffers out quicker.. but that's just me being stupid... sorry..

[Prashant Kumar_1]

We have good cpu's .. but I don't think that's the problem . It's related to the posting made by song yun keun .
http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x700aa22d6d27d5118fef0090279cd0f9,00.html

Somebody must be having an answer .. cos' it looks like a very common problem .

[harry d brown jr]

Make sure you have these patches:

PHNE_25423
PHKL_25613
PHNE_25116

and the latest java patches:
http://www.hp.com/products1/unix/java/infolibrary/patches.html


found this info, but I updated the patch list, here:
http://www.apachelabs.org/tomcat-user/200106.mbox/%3C200106271618.LAA08833@gateway.kraft.com%3E


live free or die
harry

[Prashant Kumar_1]

I've applied the three patches (PHNE_25116,PHKL_25613,PHNE_25423) you suggested mate and also reinstalled java 1.2.2.06 but it still gives the same error

HTTPClientConnection: Awaiting request on open socket.
Error HTTPListener:1
Stopped listening in HTTPListener
ListenerException: Unable to accept incoming HTTP requests.

This happens when the load balancer does the health check .

[Steven Gillard_2]

According to the explanation by Alex Glennie and others in the post you mention, ENOBUFS is now an error you must expect to be returned from accept() on hpux 11.x as it can easily be triggered by a client sending a RSET before the accept() is complete.

In our case (which was C++, not java), the solution was simply to handle the ENOBUFS as a valid error and continue normally where in the past the program would have aborted. We now simply log them as informational messages.

I imagine the situation with java is the same - until they fix the jvm to not throw an exception on this error you must catch it and continue without aborting... thus making this an application issue.

Hope this makes sense!

Regards,
Steve

[Prashant Kumar_1]

Thanks for the reply Steven . If that's the case then I'm really pissed off with HP . HP decided to change their implementation from 10.20 to 11.0 .. so the client software written for 10.20 has to be rewritten to make it work with 11.0 . How annoying is that !
I take it you are implying that the sofware has to rewritten to handle ENOBUFS . Now that is something I had not expected from an OS or a patch bundle upgrade . All we did was upgrade from 2000 patch bundle to March 2001 bundle

[Steven Gillard_2]

I can certainly understand your frustration because I got burned by this problem as well. But unfortunately that is part of the upgrading game - every time an OS upgrade or patch installation is performed something changes which causes something else to break.

The tcp stack did change considerably from hpux 10.20 - 11.0 and this is one of the side-effects.

Regards,
Steve

[Prashant Kumar_1]

But surely HP must be having some kind of fix for their jvm . Who would want to shell out more money for rewriting software to make it work with 11.0 then ?
And I thought only microsoft was capable of annoying users !