Обсуждение: index and data tablespaces on two separate drives or one RAID 0?

First and foremost, I just want to say Thank You, very much, sincerely, to
the entire PostgreSQL community for everything you are, do and have built.
Without going into an exorbitant amount of detail, suffice it to say that in
the context of my current work, you have truly been a Godsend.

I have just one immediate question that I've been unable to ascertain a
solid answer to myself by reviewing the documentation and/or looking up the
prior posts.  I suppose it might boil down to a matter of informed (and
perhaps differing) opinions, whereas for my part, the most effective way for
me to decide this on my own would be to flip a coin.  In other words, I Just
Don't Know - and the overhead of my just-trying-it-out-to-see-what-happens
would be more than I'd be able to afford.

Right now I'm running (9.1) across three physical disks:  One for table
data, one for indexes and one for WAL (in addition to the OS disk of
course).  I realize that these all should be on RAID 10s and not single
disks, and that's indeed the plan; just not a viable option for me right now
- maybe in a couple weeks...  Anyway, after an exorbitant amount of trial /
error / tweak / dumb mistake / tweak again / retry, I'm confident that my
configuration is as solid as it can be, for now - and so at present, I'm
looking just for any available avenue to improve disk I/O - seeks and reads,
specifically; writes are doing fine.

SO:  What I'm looking for is just kind of a snap consensus here among the
readership and experts:  If you were in the same position, would you either:


  A) Leave the disk setup as-is, with indexes on one drive and tables on
another?

OR

  B) Combine the two separate drives into a single RAID 0, put both indexes
and table data on that RAID, and run from there?


(the WAL drive I'll be leaving as completely separate and standalone, of
course)

With appreciation in advance for your feedback and time,

~ach

--
View this message in context:
http://postgresql.1045698.n5.nabble.com/index-and-data-tablespaces-on-two-separate-drives-or-one-RAID-0-tp5715724.html
Sent from the PostgreSQL - general mailing list archive at Nabble.com.

On 07/06/12 10:05 AM, ach wrote:
>    A) Leave the disk setup as-is, with indexes on one drive and tables on
> another?
>
> OR
>
>    B) Combine the two separate drives into a single RAID 0, put both indexes
> and table data on that RAID, and run from there?

My fastest postgresql servers have everything on one raid10, using 16 or
20 15000 rpm SAS2 drives on a 1gb flash-backed cache controller.

> (the WAL drive I'll be leaving as completely separate and standalone, of
> course)

why?

--
john r pierce                            N 37, W 122
santa cruz ca                         mid-left coast

> My fastest postgresql servers have everything on one raid10, using 16 or
> 20 15000 rpm SAS2 drives on a 1gb flash-backed cache controller.

Thank you - that affirms what'd been my own growing supposition, and the
plan

> why?

Really?  ...Well, I mean, I'd just been going with what I'd seen asserted as
the solid baseline position:  WAL should be on its own separate drive,
devoid of any interference and/or interruption other than just writing WAL.
To see that challenged is surprising; are you saying my interpretation on
that point would be incorrect, and that assumption would be wrong?

Thank you again for your feedback!

~ach

--
View this message in context:
http://postgresql.1045698.n5.nabble.com/index-and-data-tablespaces-on-two-separate-drives-or-one-RAID-0-tp5715724p5715780.html
Sent from the PostgreSQL - general mailing list archive at Nabble.com.

On 07/06/12 5:51 PM, ach wrote:
>> My fastest postgresql servers have everything on one raid10, using 16 or
>> >20 15000 rpm SAS2 drives on a 1gb flash-backed cache controller.
> Thank you - that affirms what'd been my own growing supposition, and the
> plan
>
>> >why?
> Really?  ...Well, I mean, I'd just been going with what I'd seen asserted as
> the solid baseline position:  WAL should be on its own separate drive,
> devoid of any interference and/or interruption other than just writing WAL.
> To see that challenged is surprising; are you saying my interpretation on
> that point would be incorrect, and that assumption would be wrong?

putting everything on a single large raid10 ensures the IO is evenly
distributed across all spindles, given enough workload to keep that many
concurrent IOs active.



--
john r pierce                            N 37, W 122
santa cruz ca                         mid-left coast