counters

counters

[Manik Raina]

anyone knows if there are counters in the linux kernel
which can be read via /proc like mechanism for the
following :

1. total number of bytes read by process by syscalls
like read()

2. total number of bytes written by each process by
syscalls like write()

thanks

Re: counters

[Frank Schaefer]

On Thu, 2002-05-16 at 16:14, Manik Raina wrote:
> anyone knows if there are counters in the linux kernel
> which can be read via /proc like mechanism for the
> following :
> 
> 1. total number of bytes read by process by syscalls
> like read()
> 
> 2. total number of bytes written by each process by
> syscalls like write()

Hi,

as far as I know there's not a ready to use counter in the procfs.

BTW: What do you want to count? Do You mean timers?

It shouldn't be a problem, to write a little driver, which could make
this available.

Regards
Frank

Re: counters

[Manik Raina]

	Thanks for your response. What i meant was 
	every process could have an account of how
	many bytes were read/written to various
	filesystems/sockets using read()/write()
	system calls. 

	We could dump this stuff in /proc and
	it could tell us which processes are
	heavily IO bound.

	I am wondering if this information will
	be useful to anyone.

Frank Schaefer wrote:
> 
> On Thu, 2002-05-16 at 16:14, Manik Raina wrote:
> > anyone knows if there are counters in the linux kernel
> > which can be read via /proc like mechanism for the
> > following :
> >
> > 1. total number of bytes read by process by syscalls
> > like read()
> >
> > 2. total number of bytes written by each process by
> > syscalls like write()
> 
> Hi,
> 
> as far as I know there's not a ready to use counter in the procfs.
> 
> BTW: What do you want to count? Do You mean timers?
> 
> It shouldn't be a problem, to write a little driver, which could make
> this available.
> 
> Regards
> Frank
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: counters

[Frank Schaefer]

On Fri, 2002-05-17 at 11:08, Manik Raina wrote:
> 
> 	Thanks for your response. What i meant was 
> 	every process could have an account of how
> 	many bytes were read/written to various
> 	filesystems/sockets using read()/write()
> 	system calls. 
> 
> 	We could dump this stuff in /proc and
> 	it could tell us which processes are
> 	heavily IO bound.
> 
> 	I am wondering if this information will
> 	be useful to anyone.

Hi Manik

and sorry that I read only half of your initial post.
I had a quick look at fs/read_write.c.

I don't see any hook in the functions here, to perform such a task. And
here this should belong to -- shouldn't it?

the functions could add ``count'' to the procfs entry of ``current''.
(just a thought) This info could be valuable - I think - if it wouldn't
make a large performance issue.

Regards
Frank

Re: counters

[Manik Raina]

Frank Schaefer wrote:

> Hi Manik
> 
> and sorry that I read only half of your initial post.
> I had a quick look at fs/read_write.c.
> 
> I don't see any hook in the functions here, to perform such a task. And
> here this should belong to -- shouldn't it?

	quite right. I am wondering if any locking is required for updating
	some counters which one may add in task_struct or if adding counters
	in task_struct is acceptable at all.

Re: counters

[Jerry Cooperstein]

This is doable (some other OS's do it) but:

1) It requires some changes to the basic read/write call to gather
the statistics.  It also requires stashing the counters somewhere
such as in the task_struct and thus requires modifying it.

2) It doesn't directly tell you about I/O statistics themselves
(which are available under  /proc/stat) because the I/O request
may be gotten from cache, or may never be flushed from cache
to disk depending on subsequent events, so it will always
tend to overestimate the amount of real I/O done on the device.

 Jerry Cooperstein       <coop@axian.com>
 Axian, Inc.      Software Consulting and Training
 4800 SW Griffith Dr., Ste. 202, Beaverton, OR  97005 USA
 http://www.axian.com/ 


On Thu, May 16, 2002 at 07:44:35PM +0530, Manik Raina wrote:
> anyone knows if there are counters in the linux kernel
> which can be read via /proc like mechanism for the
> following :
> 
> 1. total number of bytes read by process by syscalls
> like read()
> 
> 2. total number of bytes written by each process by
> syscalls like write()
> 
> thanks
> -

Re: counters

[Frank Schaefer]

that's right, but it would give one a picture how much a process --let's
say-- requests I/O.

Regards
Frank

On Fri, 2002-05-17 at 16:52, Jerry Cooperstein wrote:
> This is doable (some other OS's do it) but:
> 
> 1) It requires some changes to the basic read/write call to gather
> the statistics.  It also requires stashing the counters somewhere
> such as in the task_struct and thus requires modifying it.
> 
> 2) It doesn't directly tell you about I/O statistics themselves
> (which are available under  /proc/stat) because the I/O request
> may be gotten from cache, or may never be flushed from cache
> to disk depending on subsequent events, so it will always
> tend to overestimate the amount of real I/O done on the device.
> 
>  Jerry Cooperstein       <coop@axian.com>
>  Axian, Inc.      Software Consulting and Training
>  4800 SW Griffith Dr., Ste. 202, Beaverton, OR  97005 USA
>  http://www.axian.com/ 
> 
> 
> On Thu, May 16, 2002 at 07:44:35PM +0530, Manik Raina wrote:
> > anyone knows if there are counters in the linux kernel
> > which can be read via /proc like mechanism for the
> > following :
> > 
> > 1. total number of bytes read by process by syscalls
> > like read()
> > 
> > 2. total number of bytes written by each process by
> > syscalls like write()
> > 
> > thanks
> > -
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>