Re: Disk structures
Date: Mon, 22 Feb 2016 18:58:06 +0000
Message-ID: <CAFQ5AC+TUaDiDLz8p9Q-b9QRNoL2DG7E-UJj=cmv_YA95m6-TQ_at_mail.gmail.com>
Agree 100% with Kellyn. One additional thought: Not only what does your AWR say, but what do your end users say? Where are the pain points? What business processes are complained about most? That's where you should focus your attention. Is that read? Write? Both? Something else?
-Mark
On Mon, Feb 22, 2016, 13:43 Kellyn Pot'Vin-Gorman <dbakevlar_at_gmail.com> wrote:
> Hi Robert,
> I'm about to take this a different direction than answering your question,
> so bear with me.... :)
>
> You state that "scattered file reads" are you biggest waits. I don't want
> to assume the overall percentage as it wasn't stated in the email.
>
> You then go into a desire to move your redo logs on their own drives, but
> you don't state what performance data you have that shows this is a
> challenge that must be addressed with this move.
>
> You do state that you want to increase reads, but then you discuss the
> amount of audit write demands.
>
> Forgive me, but I don't see a pattern or a "story" that shows there is
> reason behind the three configuration examples.
>
> So, my question instead of an answer is: What are your top waits and the
> challenges your users are facing that you want to solve? Best practice
> means nothing if it doesn't provide value to your specific needs. If you
> run an AWR for the last week or month, (if you have the retention set to
> provide this) what does it say are your biggest challenges that can be
> addressed with a new hardware configuration/design?
>
> Thanks,
> Kellyn Pot'Vin-Gorman
> Consulting Member of Technical Staff, SCP
> Oracle
>
>
>
> [image: Kellyn Pot'Vin on about.me]
> <http://about.me/dbakevlar>
>
> Kellyn Pot'Vin-Gorman
> about.me/dbakevlar
> <http://about.me/dbakevlar>
>
> On Mon, Feb 22, 2016 at 11:31 AM, Storey, Robert (DCSO) <
> RStorey_at_dcso.nashville.org> wrote:
>
>> Sanity check. Am I overthinking / overteching the issue.
>>
>>
>>
>> Moving to a new server. Current server has the storage internal, 6
>> drives configured into a RAID 10. Overall performance is good, but, reads
>> I think could be a bit better. Most of my waits on my biggest “waits” is
>> for scattered file reads.
>>
>>
>>
>> I have 45 gig of data files and 25 gig of index files. The current
>> RAID10 holds all my files, redo, etc. Just on different logical volumnes.
>>
>>
>>
>> So, the new server has 12, 600g, 15K spin SAS drives. It also has 2,
>> 300gig 15K sas drives. My plan is to put my redo log groups on the 300 gig
>> drives, one group per drive, along with a control file copy. That way redo
>> has its own spindles.
>>
>>
>>
>> Of my 45 gig of data files, one file is 12 gig and contains a single
>> tablespace with a single table. It’s basically my audit table for all
>> actions from within the application. Every application action gets logged
>> to that table. About 25 mill rows that I keep trimmed. Has a matching 12
>> gig IDX file.
>>
>>
>>
>> I’m debating between three configs, trying to figure out what gives me
>> the better “read”. I estimate that about 70% of my system actions are
>> reads. Lots of small random writes. No bulk loads. For instance my audit
>> log gets about 65000 inserts a day. There would be another 130K or so
>> inserts to the other tables to drive that audit trail.
>>
>>
>>
>> Config A
>>
>>
>>
>> Create a RAID10 with all 12 drives. 6 pairs of RAID0 which then stripes
>> to make a 3.2tb disk pool. Logical volumes to separate the files, ie, data
>> to one volume, idx to the other.
>>
>>
>>
>> Config B
>>
>>
>>
>> Create 3 separate RAID10’s, each with 4 drives. Put all my DATA files
>> (minus the one large datafile) on one RAID 10 and IDX on the other RAID
>> 10. The Third RAID 10 would contain the separate data/idx for my largest
>> table as well as the FRA.
>>
>>
>>
>> Config C
>>
>>
>>
>> Same as B except all data to one raid, all idx to the other raid, and FRA
>> to the third array.
>>
>>
>>
>> Old school was that you wanted your data drives to not compete with the
>> index drives so that reads/writes were occurring concurrently on a data/idx
>> file.
>>
>>
>>
>> Thoughts…or am I overthinking?
>>
>>
>>
>> Thanks..
>>
>
>
-- http://www.freelists.org/webpage/oracle-lReceived on Mon Feb 22 2016 - 19:58:06 CET