Skip to Main Content
  • Questions
  • Proving / Disproving Slow IO Response from SAN

Breadcrumb

Question and Answer

Tom Kyte

Thanks for the question.

Asked: August 16, 2012 - 9:41 am UTC

Last updated: August 16, 2012 - 4:06 pm UTC

Version: 11.2.0.3.3

Viewed 1000+ times

You Asked

Tom,

Love the forum - appreciate you taking time to share your insights!

We have a situation that pops up from time to time where it seems like the IO response we're getting from our SAN is very poor. We've tried numerous methods to show the performance degradation to our Storage Admins - to 'prove our case' - but we're not having much luck. We get refuted with this 'data' or that.

Can you summarize what might be the best METHODOLOGY to use to prove (or disprove if that's the case...I'm fine with that) our case? Tracing a query - proviing IO statistics from there? Statistics from Grid? I read where you mentioned DD'ing a datafile to /dev/null...is it that simple? In all of these cases, there must be some level of 'acceptable' / 'not acceptable' - how might I determine those (I know that's subjective - but there are 'typical' values to start with before pinpointing our organizations need and capabilities)?

I'm very interested in the interaction between Oracle and the storage sub-system at this level - I'd be curious and appreciative of any information you could provide.

Thanks again!

and Tom said...

dd'ing data to /dev/null cannot be argued with.

It involves nothing about Oracle - the DBA can just step back and say "make that go fast please". The database *is not involved*.


You can also try tools such as:

http://www.oracle.com/technetwork/topics/index-089595.html

that again isn't the database - so all discussion of "it must be oracle" become moot. They aren't relevant since the database isn't involved.


Your rated IO speed and transfer rates are very personal. If you have a big bad SAN with lots of capacity and the ability to transfer 50 gigawads of data per microsecond - but you connect it to your server with a one gigabit ethernet card - you'll be *lucky* to get 100 megabytes per second.

You'd have to look at your hardware, your infrastructure - your overall architecture to determine what could be considered "good" for you.

We just issue normal OS level read and write calls.

Rating

  (1 rating)

Is this answer out of date? If it is, please let us know via a Comment

Comments

Details more on this

satheesh nagaraj, August 16, 2012 - 6:20 pm UTC

Hi Tom,

Could you please provide an example to explain on this. How to start investigating I/O issue(detailed steps, commands, interpretation of this),