MIDRANGE dot COM Mailing List Archive



Home » MIDRANGE-L » January 2014

Re: IBM i SAS drive parity sets and usable GBs - TD105880



fixed

Yes but the next failure is fatal. AND you need to have only two fail before at least one hot spare rebuild has fully completed. THEN you could have a third failure.

But because I sell disks I always configure 12 hot spares. :-)

Seriously multiple hot spares are an option, I have one large customer with dual hot spares everywhere. Never had to dip into the second one (yet!)

- Larry "DrFranken" Bolhuis

www.frankeni.com
www.iDevCloud.com
www.iInTheCloud.com

On 1/31/2014 9:15 AM, Raul A. Jager W. wrote:

RAID 6 gives you zero probability of data loss with 2 drives fail.
You use 2 extra disks, one more as hot spare and you can lose 3 drives
without data loss (as long as the spare had time to kick in)

Charles Wilt wrote:

A few years back (well ok it was almost 12 years ago :) I ran the numbers..
https://groups.google.com/forum/#!topic/comp.sys.ibm.as400.misc/_bqhgOn2KzM

"Statistically, if you have two drives fail the 2x4 RAID would give you a
57% change of having a single failed drive in each set. In other words,
if two drives fail in a 1x8 set you have 100% chance of data loss, in a
2x4 set you have only a 43% chance of data loss.
"

In one of the later posts...
"
For larger systems this doesn't seem to matter so much. For example 32

drives the chance of data loss with 2 failed drives is:
8x4-drives = 10%
4x8-drives = 23%
for 100 drives:
25x4-drives = 3%
10x10-drives = 9%

A quick conclusion is once you have to have multiple arrays because of
the number of drives, this really doesn't matter.
"

Charles




On Fri, Jan 31, 2014 at 8:14 AM, Jeff Crosby <jlcrosby@xxxxxxxxxxxxxxxx>wrote:



Very good info to know.

But, if you have 1 parity set of 24 drives, you can only lose 1 and keep
running, correct? With 3 parity sets of 8 drives, you could theoretically
lose 3 drives and still be running. I say "theoretically" because the 3
failures would have to be 1 drive in each of the 3 parity sets as opposed
to any 3 random drives.

I suppose there is a bunch of research on this as to which is the best way
to go.




On Thu, Jan 30, 2014 at 8:41 PM, Sue Baker <sue.baker@xxxxxxxxxx> wrote:



Kirk Goins <kirkgoins@xxxxxxxxx> wrote on Thu, 30 Jan 2014
20:47:05 GMT:



Per the V7R1 InfoCenter the max size of a Raid set is 18
drives. However when using a FC5913 Controller that Max is 32
per the TechDoc listed below.

Back in the days of SCSI ( vs SAS ) drives if you told the
system to build the raids sets based on Performance and you
had and there were 8 drives available to build raid it would
have built 2 sets of 4. Per the charts in the link below, the
5913 will build Performance Based raid sets of 24 drives
depending on the total number of drives and if Hot Spare is
used.

Does someone know a good doc on how Raid works in a SAS
environment? back in the SCSI days if I remember correctly (
ignore striping here ) and for a 4 drive set

The Grey is turning Greyer so this could be all wrong.
A Block of data is written to Disk #1
Controller Reads Drives #2 and #3
Calcs new Checksum and writes that to drive#4

Now with a 8 drive set
A Block of data is written to Disk #1
Controller Reads Drives #2 thru #7
Calcs new Checksum and writes that to drive#8
So there would be at least 4 more reads for that block of data
and therefor you get better performance with smaller raid
sets.

How does SAS Raid5 work differently that 24 drive raid set is
Perfomance?

TD105880
http://www-


03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/TD


105880


The doc is all about SAS drives, not SCSI drives. It is not an
exhaustive piece of documentation but is meant to help people
understand what the system will do given certain defaults.

If you want smaller parity sets than the defaults, you'll need
to implement your parity sets the "old fashioned" way by only
having the proper quantity installed when starting RAID5.

For example, if you're purchasing 24 drives and you don't want
the default but instead you want to force 3 8 drive parity sets,
don't bother with setting performance optimization. Simply make
sure only 8 drives are fully plugged in, start RAID on those.
Plug in the next set of 8, start RAID. Plug in the final set of
8, start RAID. However, you will not see the best performance
because there will be 2 parity sets aligned to one adapter and 1
parity set aligned to the 2nd adapter in the pair.

RAID5 works the same regardless of SCSI or SAS. For each write
to a drive, 4 ops must occur at the disk level. Read where the
data will go, read the parity stripe, write the parity stripe,
write the data.

The only time 4 drives per parity set is the best choice is when
you plan to operate with broken drives. When operating with
fully functioning drives, the more drives that can participate
in the parity set the better to avoid disk waits and hot spots.

--
Sue
IBM Americas Advanced Technical Skills (ATS) Power Systems
Rochester, MN
--
This is the Midrange Systems Technical Discussion (MIDRANGE-L) mailing


list


To post a message email: MIDRANGE-L@xxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/mailman/listinfo/midrange-l
or email: MIDRANGE-L-request@xxxxxxxxxxxx
Before posting, please take a moment to review the archives
at http://archive.midrange.com/midrange-l.




--
Jeff Crosby
VP Information Systems
UniPro FoodService/Dilgard
P.O. Box 13369
Ft. Wayne, IN 46868-3369
260-422-7531
www.dilgardfoods.com

The opinions expressed are my own and not necessarily the opinion of my
company. Unless I say so.
--
This is the Midrange Systems Technical Discussion (MIDRANGE-L) mailing list
To post a message email: MIDRANGE-L@xxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/mailman/listinfo/midrange-l
or email: MIDRANGE-L-request@xxxxxxxxxxxx
Before posting, please take a moment to review the archives
at http://archive.midrange.com/midrange-l.










Return to Archive home page | Return to MIDRANGE.COM home page

This mailing list archive is Copyright 1997-2014 by MIDRANGE dot COM and David Gibbs as a compilation work. Use of the archive is restricted to research of a business or technical nature. Any other uses are prohibited. Full details are available here. If you have questions about this, please contact