Hello

I am assuming that you are refering to Short Stroking - a method to increase performance by using only a sub-set of available tracks - typically the drive's outer tracks..

As far as placement of data on a disk - you have no control over that. Also there is no method to control this.

Also as the Clariion is a cached array there should be no reason to do this.

By not using the rest of the Raid Group for some other purpose you are eliminating contention and so potentially not reducing the performance for the volume created. However as this is a cached array - this will be minimal.

I hope this helps.

Sam Claret TSE3

short stroking is a way to "program" a drive in such a way that the drive is for example a true 146GB disk and after the reprogram you can only see 73GB on the outside of the drive. A Storage Array will never know it's in fact a larger drive.

But the way you describe it sounds valid. In the performance workshop I was told that the first LUN will indeed always be created on the fastest track. If you make sure you only fill it up for a certin amount of GBs and make sure the disk is defragged, you could call this a form of software short stroking. In a way the IOps a drive can deliver are now delivered to (for example) only half the frive, since it's half full, but the number of IOps stays the same, so the number of IOps per GB is higher !

Performance can be increased by binding LUNs not using the full capacity of the physical RAID group.  This leaves some of the RAID group’s capacity unused in a technique called short-stroking. The short stroking technique is becoming less common with the introduction of flash drives, which can produce very high IOPs in a 'short' RAID group.

As previously mentioned, the first LUN bound to a RAID group is on the outermost tracks.

Short-stroking is discussed in several sections of EMC CLARiiON Storage System Fundamentals whitepaper, for example in the 'LUN Performance' section.  The topology of LUN placement on RAID groups is also discussed in that document. There is also a discussion of the I/O types that benefit most from short-stroking and alternative techniques.   Fundamentals is available on Powerlink.

I wish

I wish I could even get my hands on any VNX.....

Hello

It is not control as such - you can not specify to start at a particular offset in the RG on the disks.

I suppose if you wanted to leave a specific space at the beginning of a disk you can create a small LUN and then the next LUN would start after that.

If you have an empty RG with a set of spindles in it and you create a LUN stripped across the spindles then the LUN will be built on the outer edge of the platters growing inwards. Personally I dont see this as control - more taking advantage of the way the Clariion works.

Sam Claret EMC TSE3

That's why we would choose as much small drives as possible instead of just a few huge drives. More is better !

But even better: I want EFD's.....

Ok thanks for all your imput. Just to confirm - your collegue says that there is no way of controlling data placement on disk (tracks) but you say that when the first Lun is created on a RG the Lun is created on the outer edge - is this not data on disk control?

With flash drives, 'less is more'.  However, 'going wide' with the RAID group is the best alternative to short stroking mechanical drives, and is a more efficient use of capacity.

RRR, I assumed you have a whole DAE of flash drives on that new VNX 7500, you've got on order.

I was actually interested in this as my Boss insists that Short-Stroking is the way to go.  As a result, when I recieved my new CX4-480 a few weeks back, I actually tested this by creating new spindles in empty RGs.  The array was connected with 8GB fiber to the host on I/O empty fiber switches, so the only traffic to or from the array was for the host with the attached storage.  I tried two setups.

1 - Create new RG and new LUN on the outside (first LUN) of the spindles on a 8 Spindle Raid 1/0 set.

2 - Created new RG and new 300GB LUN and then new 300GB LUN (outside of disk spindles) on a simmilar 8 Spindles Raid 1/0 set.  I then removed the original 300GB LUN.  This left the used LUN on the outside.

I then used IO Meter to test the performance of each configured LUN.  Due to the cache on the array I was unable to notice any difference in performance.  I would say that at this point in time Short-Stroking is no longer needed unless you are in Symetrix world.

TJ