Amos
That is a great answer for network connections. Thanks - I am familiar
with netstat -p
My question regards DISK IO, (remember when processes were said to be
io-bound or cpu-bound?)
I have a simply question, namely How many disk block reads and writes
is a process doing over a given
Danny Lieberman wrote:
I have a simply question, namely How many disk block reads and writes
is a process doing over a given period of time, and what is the minimum,
maximum and average stats
This seemingly trivial question (which can easily be answered in a
Windows operating system using
I think that the gprof output and equivalent is the furthest one can go
to appreciate IO stats. I simply measured disk system throughput (using
hdparm etc) and then tuned my programs until they ran almost that fast,
then stopped worrying. For other types of IO the maximum bandwidth can
be
Peter
As a general approach for tuning a complex, working system, the first
step is to identify bottlenecks under real load, that includes both IO
rates per process and CPU rate per process.
Your method is good for a static system where you have complete control
of the applications - I am not
do it the io-accounting way -
i.e. count cache io) - the numbers will be skewed, but they will be
skewed for everyone.
I would settle for a io stats per process (similar to sar, perhaps)
for a sampling period (to be defined): current, moving average,
maximum, minimum.,
For chrissake - my
On Wed, 27 Dec 2006, Danny Lieberman wrote:
Peter
As a general approach for tuning a complex, working system, the first step is
to identify bottlenecks under real load, that includes both IO rates per
process and CPU rate per process.
Your method is good for a static system where you have
Gilad
One more input, which is important (imho) - this is a performance
measurement / system tuning objective, not a system accounting
application - i.e.we are not trying to charge someone
for io usage ( i think they used to do that once upon a time).
10x
dL
Gilad Ben-Yossef wrote:
I think that you should take a good look at the gperf etc output. (even
time(bash)) gives some hints (e.g. system time of a program ~= IO load
it produces, if IO is a major part of the runtime).
Peter
=
To unsubscribe, send
On Wed, Dec 27, 2006 at 04:31:02PM +0200, Danny Lieberman wrote:
Let's say (for the sake of argument, we do it the io-accounting way
- i.e. count cache io) - the numbers will be skewed, but they will
be skewed for everyone.
I would settle for a io stats per process (similar to sar, perhaps
stats per process (similar to sar, perhaps)
for a sampling period (to be defined): current, moving average,
maximum, minimum.,
For chrissake - my road bike speedometer can do that - surely we can
figure out a way to do it in Linux ;-)
With the caeats Gilad mentioned and considering the fact
On 22/12/06, Beni Cherniavsky [EMAIL PROTECTED] wrote:
On 20/12/06, Danny Lieberman [EMAIL PROTECTED] wrote:
How do I get the real IO (block reads/writes per second, not cached) of
each process on a running Linux system?
vmstat and iostat dont provide process level detail
Same question
On 20/12/06, Danny Lieberman [EMAIL PROTECTED] wrote:
How do I get the real IO (block reads/writes per second, not cached) of
each process on a running Linux system?
vmstat and iostat dont provide process level detail
Same question about sockets - how do I find out which process hogs my
guys
How do I get the real IO (block reads/writes per second, not cached) of
each process on a running Linux system?
vmstat and iostat dont provide process level detail
I'm looking at a system right now which is CPU idle but very IO busy and
I cant figure out who's moving the disks around.
Check out the archives.
We discussed this in the past.
IIRC, you can use echo 1 /proc/sys/vm/block_dump
(it might also involve /proc/sys/vm/laptop_mode but i can't remember why).
and then tail /proc/kmsg to see the dump of the processes i/o requests.
On Wednesday 20 December 2006 12:32, Danny
Tzahi
Thx - Fair enough - I can look in archives.
I may be mistaken, but afaik, there is no /proc/sys/vm/block_dump on Red
Hat EL3 - I couldn't find any /proc counters that were per process and
all the counters I can see
are system-wide.
Looking for a real solution today; I started poking
15 matches
Mail list logo
