setting up 3 x 12H's

[Anwar Sukkar]

Hi All

I have 3 x 12H array 2 have 12 x 18Gb disk modules and the 3rd has 6 x 18Gb. They all have dual controllers, I have to connect them to a K570, now my problem is I need performance so should I daisy chain the 3 arrays to 2x HBAs or just connect each array to its own HBA? also is there any detailed documentation on setting up a 12H for maximum perormance under SAP & Oracel. Thanks

Anwar

[Trevor Roddam_1]

Anwar.

Very interesting question.
There have been libraries of books written on just this senario.

I've included a HP White paper on performance and the 12H. It is a bit heavy but worth the read.

Essentially I would look at Seperate pairs of HBA's for each 12H if possible.

I wouldn't have more than eight LUN's on any SCSI bus. This is a, general, break even point for performance on FW SCSI.
There are two basic approaches:
1/ generate equal sized LUN's accross multiple 12H's and stripe them with alternating Pri/Alt paths to ge the best from the 12H's and the SCSI paths.
2/ manage the 12H's to get the best out of each mount point.

I pref the former, using multiple SCSI cards. This is more complicated and restricts some of the nicer benifits of 12H's. It also doubles up on over head but can if done correctly improve performance.

The second method is easier t impliment and will generally give a better perfomance, because it is easier to maintain/impliment.

So what do you do?
I would concentrate on getting the most from the "write working set"s. In otherwords only allocate the minimum from each 12H over 48% of physical to maximise their performance. Concentrate on getting performance from the two fully populated ones and use the third as the dumping ground for binaries etc.

To get the best from a mount point it should have ~6 LUNs supporting it. 4 to eight is OK.
So
Start at 2 LUNs from the 12H to make a single mount point and have the mount point spriped over 2 arrays, remember to alternate pri/alt paths to the disks.

Have ~6 VG's to cover the 2 12H's. This has 6 pri paths to mount points per HBA giving room for growth of current Lvol's by adding more LUN's or creating new lvols.
Having two lvol's per VG striped evenly over all the LUNS available. Have it all in one VG if you wuold like, I used them to split the config up.

Allocate hot spares and keep allocated space to less than 60% of physical, pref 48%.

Use the 3rd 12H in a similar manner over its own HBA's to provide /opt space for binaries, scripts etc.

I had several links to documents that have dissappeared as a result of the product going obsolete.

If you would like a drawing of this config let me know.
BTW: this is but one method, there are if anything too many.

[Anwar Sukkar]

Thank you for your reply Trevor, what I think I will do is connect each array to its individual HBA but in doing it this way do you think I should link the x y controllers together or not what are the pros and cons ...., I read somewhere that if a controller is not phiscally connected to a HBA its being used.

[A. Clay Stephenson · Accepted solution]

Dedicating one HBA per 12H is the least desirable option you have. If you lose an HBA, terminator, or cable the entire 12H is dead. The onlt thing you are guarding against is the failure on an AutoRAID controller. Your better option is to make sure that each 12H is connected to two HBA's. You can set it up so that most of the traffic to one of the 12H goes thru HBA0 and most of the traffic to the other 12H goes thru HBA1.

I do not agree that there is any reason to split a mountpoint/filesystem to any more than 2 identicals LUN's; I have never been able to measure any throughput improvement beyond 2 LUN's per filesystem striping each LVOL at 64K.
Going to more than 2 LUN's per mountpoint only makes things APPEAR better to tools like Glance because all it knows is that a tremendous amount of I/O is going thru a device. The ACTUAL throughput is the same and since you can only have 8 LUN's per AutoRAID, why waste them?

I would cable something like this:

HBA0 ----- 12H-1X ------ 12H-2Y -- Term.
HBA1 ----- 12H-2X ------ 12H-3Y -- Term.
HBA2 ----- 12H-3X ------ 12H-1Y -- Term.

Each filesystems should be comprised of 2 identical LUNS, LUN0 should be thru primary path X (alternate y); LUN1 should be thru primary path Y (alternate X). When you create each LVOL stripe across both LUN's in 64k chunks. The above scheme will give you optimum performance and redundancy. As has already been said, leave as much space as possible unconfigued on each 12H so that it stays in RAID 1/0 mode at all times.