Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by NetComra » Wed, 01 Sep 1999 04:00:00



Hi folks,

We are thinking to place one of our DB's on a Sun Storedge 5200 and
Sun Enterprise 4500.

Since we have a web-based OLTP environment, well, actually mixed,
since we have major updates running at nights against most of the
data, I thought that RAID-5 is not great, since during the day there
are a lot of small reads/writes.
I was thinking about 0+1 on datafiles with actualy data, but I am not
sure whether I need to mirror the TEMP tablespace or whether I should
place the REDO logs on the mirror.

Any advice would be appreciated.

Thanx
---------------
Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
AOL: NetComrade  ICQ: 11340726 remove NSPAM to email

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by Martin Hepwort » Thu, 02 Sep 1999 04:00:00



> Hi folks,

> We are thinking to place one of our DB's on a Sun Storedge 5200 and
> Sun Enterprise 4500.

> Since we have a web-based OLTP environment, well, actually mixed,
> since we have major updates running at nights against most of the
> data, I thought that RAID-5 is not great, since during the day there
> are a lot of small reads/writes.
> I was thinking about 0+1 on datafiles with actualy data, but I am not
> sure whether I need to mirror the TEMP tablespace or whether I should
> place the REDO logs on the mirror.

> Any advice would be appreciated.

> Thanx
> ---------------
> Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
> AOL: NetComrade  ICQ: 11340726 remove NSPAM to email

HI
Well I'm doing similar and am about to test various combinations...right
now I'm going for 2x RAID5 sets, 1 for the data and 1 for the redo logs.
With the 5200 you have hardware RAID5 so writing should be nice and
fast, not that much different from 0+1, and RAID 0+1 is probably not
justifiable for extra tiny perf gain agianst cost of disks.

But like I said I'm going to test a few combinations and see what
happens. If you want to pool test results contact me via email.

Martin

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by Bob Fazi » Thu, 02 Sep 1999 04:00:00


I would like to hear your results using raid 5 for redo logs.

As far as should you or shouldn't you mirror temp.  The question can only
be, can you afford to have the system crash without warning.  Is so, you can
probably get by without mirroring temp or rollback.   You just may crash.



> > Hi folks,

> > We are thinking to place one of our DB's on a Sun Storedge 5200 and
> > Sun Enterprise 4500.

> > Since we have a web-based OLTP environment, well, actually mixed,
> > since we have major updates running at nights against most of the
> > data, I thought that RAID-5 is not great, since during the day there
> > are a lot of small reads/writes.
> > I was thinking about 0+1 on datafiles with actualy data, but I am not
> > sure whether I need to mirror the TEMP tablespace or whether I should
> > place the REDO logs on the mirror.

> > Any advice would be appreciated.

> > Thanx
> > ---------------
> > Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
> > AOL: NetComrade  ICQ: 11340726 remove NSPAM to email

> HI
> Well I'm doing similar and am about to test various combinations...right
> now I'm going for 2x RAID5 sets, 1 for the data and 1 for the redo logs.
> With the 5200 you have hardware RAID5 so writing should be nice and
> fast, not that much different from 0+1, and RAID 0+1 is probably not
> justifiable for extra tiny perf gain agianst cost of disks.

> But like I said I'm going to test a few combinations and see what
> happens. If you want to pool test results contact me via email.

> Martin

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by davidr.. » Thu, 02 Sep 1999 04:00:00


Quote:> Since we have a web-based OLTP environment, well, actually mixed,
> since we have major updates running at nights against most of the
> data, I thought that RAID-5 is not great, since during the day there
> are a lot of small reads/writes.
> I was thinking about 0+1 on datafiles with actualy data, but I am not
> sure whether I need to mirror the TEMP tablespace or whether I should
> place the REDO logs on the mirror.

Check out this site.  It talks about some NT specific things, but the
ideas about RAID are generic enough for translation to whatever
platform desired.

http://www.ipass.net/~davesisk/oont_performance.htm

-David Rice

Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by Andrew L. Garma » Thu, 02 Sep 1999 04:00:00


Quote:> With the 5200 you have hardware RAID5 so writing should be nice and
> fast, not that much different from 0+1, and RAID 0+1 is probably not
> justifiable for extra tiny perf gain agianst cost of disks.

The A5200 does not have hardware RAID5.  None of the A5x00 series have
hardware RAID.  The closest you can get is using the Fast Write Cache Card
that has 32 MB of mirrored NVRAM on it.  I have to say, in large
installations, that card doesn't have enough NVRAM.

If you want to do hardware RAID with Sun products, the A1000, A3500 (with
D1000 in the cabinet), A7000 and Sun's PCI RAID controller for E250 and
E450 are the only options.  The A5x00 series don't work well with RAID-5!
The D1000 (standalone, without the A3500) is even worse!

Regards,

Andrew Garman
Goliath Networks, inc.
a Sun Elite Reseller &
IBM Storage Partner  :)

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by Jeff Hunte » Thu, 02 Sep 1999 04:00:00


Just FYI, the A5200 is not a Hardware based RAID unit; it is all software
based RAID...


>> Hi folks,

>> We are thinking to place one of our DB's on a Sun Storedge 5200 and
>> Sun Enterprise 4500.

>> Since we have a web-based OLTP environment, well, actually mixed,
>> since we have major updates running at nights against most of the
>> data, I thought that RAID-5 is not great, since during the day there
>> are a lot of small reads/writes.
>> I was thinking about 0+1 on datafiles with actualy data, but I am not
>> sure whether I need to mirror the TEMP tablespace or whether I should
>> place the REDO logs on the mirror.

>> Any advice would be appreciated.

>> Thanx
>> ---------------
>> Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
>> AOL: NetComrade  ICQ: 11340726 remove NSPAM to email

>HI
>Well I'm doing similar and am about to test various combinations...right
>now I'm going for 2x RAID5 sets, 1 for the data and 1 for the redo logs.
>With the 5200 you have hardware RAID5 so writing should be nice and
>fast, not that much different from 0+1, and RAID 0+1 is probably not
>justifiable for extra tiny perf gain agianst cost of disks.

>But like I said I'm going to test a few combinations and see what
>happens. If you want to pool test results contact me via email.

>Martin

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by dougc.. » Thu, 02 Sep 1999 04:00:00




Quote:> Hi folks,

> We are thinking to place one of our DB's on a Sun Storedge 5200 and
> Sun Enterprise 4500.

> Since we have a web-based OLTP environment, well, actually mixed,
> since we have major updates running at nights against most of the
> data, I thought that RAID-5 is not great, since during the day there
> are a lot of small reads/writes.
> I was thinking about 0+1 on datafiles with actualy data, but I am not
> sure whether I need to mirror the TEMP tablespace or whether I should
> place the REDO logs on the mirror.

> Any advice would be appreciated.

> Thanx
> ---------------
> Andrey Dmitriev     eFax: (978) 383-5892  Daytime: (917) 373-5417
> AOL: NetComrade     ICQ: 11340726 remove NSPAM to email

I have done many many many tests of RAID 5 versuses RAID 1 versus RAID
0+1 (or Raid Ten as I call it). Personally, I have yet to see RAID 10
not offer SIGNIFICANT performance gains in both read only and especially
in large update environments. I have used vendor DASD and shared DASD
such as EMC. Even with the big EMC cache, RAID 5 smells for a DB server.

The MINUMIM I have seen is a 30% gain in a read only environment on
some selected queries when the moon was in the right phase. I have seen
it go as much as 1800%. I had to redo this one several times to validate
my numbers because even I did not belive this dramatic a gain, but i
did it several times. (This was in a Sybase world)

Lots of people use RAID 5 and think it's great and I can envision that
there are a small percentage of cases where it will do ok, but..... For
the price of today's DASD why cut corners and create unnecessary
headaches for yourself and your customers? Do it right the first time
and go RAID 10 all the way.  IMHO.

Your milage may vary.... Use a performance tool to see for yourself. If
your reads are 5-8ms and writes are 5-12ms you are in great shape.
However if you are seeing 20, 30 , 50 , 100, 500, or (i saw this) 1500
ms, change to something else.

Also
- If you are mirroring, there is really no reason for multiple redo log
members in the groups.
- But archive redo on fast drives.

Good Luck

Doug Coan
Senior Client Server Systems Integrator
Aegon USA

Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by bcho.. » Thu, 02 Sep 1999 04:00:00


I have only one comment regarding Doug's statement below on
mirroring and multiplexing redos.

Mirroring redo and multiplexing redo log members are 2 different
things. We had a real life experience that a DBA mistakenly
removed all redo logs while system is running.
In this case mirror would not help a bit.

But another member would have saved our butts - sort like
you have multiple controlfiles.

Mirror also mirrors human mistakes, but multi-members won't.

So we do both since then.

-Bass Chorng

Quote:> Also
> - If you are mirroring, there is really no reason for multiple redo log
> members in the groups.
> - But archive redo on fast drives.

> Good Luck

> Doug Coan
> Senior Client Server Systems Integrator
> Aegon USA

> Sent via Deja.com http://www.deja.com/
> Share what you know. Learn what you don't.

Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.
 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by David Sis » Thu, 02 Sep 1999 04:00:00


Hi:

See the article on RAID on the site below (Performance -> RAID).  It's
specific to NT, but the rules are the same for Unix.

Regards,

--
David C. Sisk
Need tech info on Oracle? Visit The Unofficial ORACLE on NT site at
http://www.ipass.net/~davesisk/oont.htm
Like original modern rock?  Listen to song samples and buy a CD at
http://www.mp3.com/disparityofcult



>> Hi folks,

>> We are thinking to place one of our DB's on a Sun Storedge 5200 and
>> Sun Enterprise 4500.

>> Since we have a web-based OLTP environment, well, actually mixed,
>> since we have major updates running at nights against most of the
>> data, I thought that RAID-5 is not great, since during the day there
>> are a lot of small reads/writes.
>> I was thinking about 0+1 on datafiles with actualy data, but I am not
>> sure whether I need to mirror the TEMP tablespace or whether I should
>> place the REDO logs on the mirror.

>> Any advice would be appreciated.

>> Thanx
>> ---------------
>> Andrey Dmitriev  eFax: (978) 383-5892  Daytime: (917) 373-5417
>> AOL: NetComrade  ICQ: 11340726 remove NSPAM to email

>HI
>Well I'm doing similar and am about to test various combinations...right
>now I'm going for 2x RAID5 sets, 1 for the data and 1 for the redo logs.
>With the 5200 you have hardware RAID5 so writing should be nice and
>fast, not that much different from 0+1, and RAID 0+1 is probably not
>justifiable for extra tiny perf gain agianst cost of disks.

>But like I said I'm going to test a few combinations and see what
>happens. If you want to pool test results contact me via email.

>Martin

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by Martin Hepwort » Fri, 03 Sep 1999 04:00:00



> > With the 5200 you have hardware RAID5 so writing should be nice and
> > fast, not that much different from 0+1, and RAID 0+1 is probably not
> > justifiable for extra tiny perf gain agianst cost of disks.

> The A5200 does not have hardware RAID5.  None of the A5x00 series have
> hardware RAID.  The closest you can get is using the Fast Write Cache Card
> that has 32 MB of mirrored NVRAM on it.  I have to say, in large
> installations, that card doesn't have enough NVRAM.

> If you want to do hardware RAID with Sun products, the A1000, A3500 (with
> D1000 in the cabinet), A7000 and Sun's PCI RAID controller for E250 and
> E450 are the only options.  The A5x00 series don't work well with RAID-5!
> The D1000 (standalone, without the A3500) is even worse!

> Regards,

> Andrew Garman
> Goliath Networks, inc.
> a Sun Elite Reseller &
> IBM Storage Partner  :)

I stand corrected. We've got A1000's here for precisely the reason that
it does hardware RAID. In this case you are 'stuck' with RAID 0+1. But
agian testing of placing the various files in various combination for
you specific application is the way forward.

Martin

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by dougc.. » Fri, 03 Sep 1999 04:00:00




> I have only one comment regarding Doug's statement below on
> mirroring and multiplexing redos.

> Mirroring redo and multiplexing redo log members are 2 different
> things. We had a real life experience that a DBA mistakenly
> removed all redo logs while system is running.
> In this case mirror would not help a bit.

> But another member would have saved our butts - sort like
> you have multiple controlfiles.

> Mirror also mirrors human mistakes, but multi-members won't.

Excellent point! That is the only area where we thought the mirror would
not help.  I have heard of one other place that did this (erased
the redos).  Hopefully all turned out ok for you.

Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by Andrew L. Garma » Fri, 03 Sep 1999 04:00:00


Broad generalizations don't work out in the field.  RAID 0+1 performs
better than RAID-5 in all cases I've ever seen, but that's not a
sufficient reason to use it.  Oracle does an excellent job of caching
it's tables.  Most environments I work in there is between 4GB and 24GB of
RAM available.  It's quite acceptable to put the Tables and application on
a RAID-5 volume because all reads are going to be from cache.

In the same regard, putting Redo, Rollbacks or temp Table Space on RAID-5
will kill your performance.  Even with hardware RAID!  So RAID 0+1 are the
only RAID option I can recommend.

But as always, your mileage may vary.  Take a baseline measurement, using
'iostat -xn' or 'vxstat'.  Look for disk a_svct and b% that are measured
in the 10's.  If you find either, you need more disks, or to rethink how
your disks are configured.

Regards,

Andrew Garman
Goliath Networks, inc.

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by Andrew L. Garma » Fri, 03 Sep 1999 04:00:00



> We are thinking to place one of our DB's on a Sun Storedge 5200 and
> Sun Enterprise 4500.

> Since we have a web-based OLTP environment, well, actually mixed,
> since we have major updates running at nights against most of the
> data, I thought that RAID-5 is not great, since during the day there
> are a lot of small reads/writes.
> I was thinking about 0+1 on datafiles with actualy data, but I am not
> sure whether I need to mirror the TEMP tablespace or whether I should
> place the REDO logs on the mirror.

If you can have a job abruptly stop when the disk TEMP tablespace is on,
then you don't need to mirror it.

For the A5200, I'd recommend you buy two.  And add a second I/O board to
the E4500.  Buy the four additional GBICs and mirror between the two
arrays.  Each array should have a connection from both I/O boards.  This
way you can lose any I/O component and still function!  The performance is
also phenomenal!  Set the stripe width to 32 and away you go!!

Just my $.02,

Andrew Garman
Goliath Networks, inc.

 
 
 

Raid performace issues (Raid 5 vs. Raid 0+1) with database files

Post by Anthony Mandi » Mon, 06 Sep 1999 04:00:00



> The A5200 does not have hardware RAID5.  None of the A5x00 series have
> hardware RAID.  The closest you can get is using the Fast Write Cache Card
> that has 32 MB of mirrored NVRAM on it.  I have to say, in large
> installations, that card doesn't have enough NVRAM.

        Mirrored NVRAM?!?? What a concept!

-am

 
 
 

1. raid 0+1 or raid 5+1?

Hello,

The magic question,  In setting up an Oracle based OLTP app, which
should I use, raid 0+1 or raid 5+1 (Assume hw based raid 5 and mirroring
via ods or vxvm).

Note:  I am looking for the right balance between price, performance and
reliability.

Any comments are welcome.

Thanks,

Ron

--
+--------------------------------------------------------+
|   Ron Dilley                  Sr. UNIX Administrator   |

|   Amgen Inc.                           (805)447-6730   |
+--------------------------------------------------------+

2. problems with mprotect()

3. RAID 0+1 vs RAID 5

4. Home URL does not work in Konqueror

5. Veritas's implementation/interpretation of RAID 0+1 and RAID 10

6. Building High Performance Linux Internet Server!

7. Raid 1+0 or Raid 0+1

8. Cachefs, CDROM and nfs mount question

9. Help Configuring RAID 0+1 using RAID Manager 6.22

10. RAID RAID RAID

11. HW RAID vs SW RAID

12. RS/6000 RAID vs. AIX RAID

13. IDE RAID Cards vs Software RAID????