fork problems

[Marcin Kurc]

Hi,
I'm running software which produces scratch files. Every time the file size reaches 32768 k I'm getting the following error:
*** Operating System Function -
*** fork() failed : Not enough space (errno=12)

Could anybody tell me what kernel parameter is responsible for this behavior?
thank you

[Justo Exposito]

Hi,

I think that is nproc, if you have glance you can view it. If not there are a evaluation copy in the applications CD.

Regards,

Justo.

[harry d brown jr]

maxdsiz

Post what you have from kmtune

live free or die
harry

[Marcin Kurc]

here is my kmtune output attached

[James R. Ferguson]

Hi Marcin:

It is very likely that you have insufficient swap space. Have a look via:

# swapinfo -tam

Regards!

...JRF...

[John Palmer]

Given the following from 'man fork'...

[ENOMEM] There is insufficient swap space and/or physical memory available in which to create the new process.

(ENOMEM is error 12).

I'm not convinced that a kernel parameter is responsible. Is the application issuing lots of forks?

Regards,
John

[Marcin Kurc]

Thanks everybody,
Jamses, there seems to be enough swap space.
Yes, the application uses a lot of forks. Also the application does not use all the memory. I thought that it might have somthing to do with the kernel limits because of the specific size of the scratch file while it crashes.

[Rita C Workman]

Your parms list is interesting..
Jim mentions looking at swap..I'd add look at how your parms are handling swap.
You have bufpages set to the calc of nbuf*2...nbuf is set to default of 0 (I would have set bufpages to 0 as well, so these could work dynamically to control buffer).
You have swapmem turned off=0 (I would have turned it on, since I like the system to control memory dynamically).

What I'm saying is looking at your parms....it says your trying to set up to control that buffer stuff(now Jim & Clay can do that - I'm not that good...so I let the system handle as much of this as I can set up..)

So what I'm saying is - look at how much swap you have set up...and set up your parms to allow for more dynamic control of cache...
I only looked at this point of view-there were some other things in your parms that caught my eye...but not knowing your system or apps, I felt best to leave those alone.

I haven't finished my coffee-so hope this made some sense.

Rgrds,
Rit

[Marcin Kurc]

Thanks Rita, I'll try it.
This box is optimized for CAD/CAM.
Everything makes sense, except bufpages if nbuf=0 and bufpages=NBUF*2 then bufpages=0
I had 3 coffees this morning :)

[James R. Ferguson]

Hi Marcin:

Please post the output of 'swapinfo -tam'. As Rita noted, you have pseudoswap disabled. This means that you *must* have an equal amount of swap as you do physical memory in order to start a process. This could be very limiting for you!

Regards!

...JRF...

[Marcin Kurc]

Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 1024 61 963 6% 0 - 1 /dev/vg00/lvol2
reserve - 380 -380
total 1024 441 583 43% - 0 -


Also, this box has 1024 megs RAM and it's using around 500 on average.

[James R. Ferguson]

Hi (again) Marcin:

If nbuf=0 and bufpages=0 then you have enabled the dynamic buffer cache, and hence dbc_max_pct and dbc_min_pct operate. Your values for those are quite convervative. The buffer cache speeds file I/O by leaving file's buffer's in memory. I don't think this is a problem here.

Regards!

...JRF...

[Marcin Kurc]

James,
dbc_max_pct and dbc_min_pct values were suggested by tech support (app developers) it didn't do much good anyways.

[Rita C Workman]

Marcin,

First tip...listen but not necessarily comply with a vendor's stmt on what your parms should be.

You need to watch and determine how the system should be tuned....

I agree w/Jim those dbc_max% and dbc_min% are really pinching the penny...

Just a thought...coffee in hand..
Rita

[oiram]

Hi,


Perhaps this is too obvious but have you checked your ulimit values?

Best regards.

[A. Clay Stephenson]

Hi:

I've had quite a bit of experience with CAD/CAM/Analysis packages and many of them do funny/stupid things with the .profile (or more often .cshrc). After looking at this, I'm beginning to suspect that we are overlooking the obvious - Soft limits. I would do a ulimit -a. Also, just so we know for sure, a few ps's to display memory usage would be good along with some swapinfo -a output.

Use this Bill H. variant of ps to determine memory usage:
UNIX95= ps -e -o vsz,ruser,pid,args

If you will post that stuff, I think that this can be nailed down. Of course, Glance would be much better but ...

Good hunting, Clay

[Marcin Kurc]

Here are my ulimit settings
time(seconds) unlimited
file(blocks) unlimited
data(kbytes) unlimited
stack(kbytes) 81612
memory(kbytes) unlimited
coredump(blocks) 4194303
nofiles(descriptors) 512

and ps output attached

[A. Clay Stephenson · Accepted solution]

HI:

You have a 725MB process running on a 1000MB box. When that forks, you are dogmeat. Your average memory usage of around 50% matters not, it's the peaks that are killing you. You are one of those rare modern cases that need 2-3x swapspace and/or a lot more memory. If you don't have any raw lvol space left, you might add filesystem swap at low priority but you really need more memory.

[Marcin Kurc]

Thanks Clay,
I'll try to add more swap, if that doesn't work then I'll have to steel some memory from those surfing the internet.

[James R. Ferguson]

Hi Marcin:

Since adding additional swapspace (given that you have adequate 'maxswapchunks' based on you 'kmtune' output) doesn't require a reboot, you could do that.

In addition, I think I'd schedule a kernel regeneration and enable pseudoswap (swapmem_on=1). As I alluded in an earlier post, this will allow more processes to start since process space isn't reserved for three fourths of physical memory when psuedoswap is enabled.

The small overhead associated with pseudoswap is probably less important to you than anything else.

Regards!

...JRF...

[Marcin Kurc]

Thank you everyone for your help and comments.

[Kevin_31]

Can you update on what you ended up finding/doing and whether you know which action solved your problem (or if it's still ongoing?

I've got a similar problem and am very curious.

cheers,
Kevin