Unsolved
This post is more than 5 years old
2 Posts
0
25249
November 22nd, 2011 08:00
PERFORMANCE - "Short Stroking" - balancing different drive capacities in the same tier
Hi everyone
Here's a problem we have run into, and probably affects a number of you too.
The issue is simple - disk drive sizes keep increasing, and it is hard to keep buying the same size of disks for each Tier of storage in your Compellent Storage Center. For example, in our system, we originally purchased 300GB Fibre Channel disks. Compellent stopped selling these in October 2010, so we installed 450GB SAS drives instead.
This means that you have some choices to make about adding new disks which are a different size (larger) to the old ones:
(1) Add the disks as a completely separate disk folder. This means that you are now managing two pools of storage.
(2) Add the disks to the same storage pool.
From an administrative workload as well as space utilisation efficiency point of view, (2) is a better choice.
However, option (2) does present some problems.
If a system has two types of disk, then at some point the smaller disks will fill up, and some data will necessarily be striped across the larger disks and NOT the smaller ones. This effectively results in uneven balancing of the I/O load. The effect is more pronounced in systems which were more full before the disk upgrade took place, since the restripe algorithm gets less efficient the more full the system is already (apparently).
In practices, this means that the system COULD start to experience hot spots where there is an imbalance of drive sizes.
I've discussed this issue with various people at Dell Compellent, and the only obvious solutions to this are option (1) above, or to "short stroke" the new drives.
What do I mean by short stroking in this context?
In other high-end systems, this means confining the disk read head to a particular area of the disk, so that the read head never performs a "full seek stroke" - thus minimising latency. The side effect of this is also to restrict the usable capacity of the disk. In this case, we want to restrict the capacity of the 450GB disks so that they only have 300GB accessible - this will hopefully mean better balancing of I/O across both old and new disks.
So, thanks for your patience in reading this post... please vote below to let me know your views on "short stroking" in the Compellent SAN!
[Poll]
Samnadeau
11 Posts
0
November 22nd, 2011 17:00
Yep you're right..its a known limitation..and it's understandable given how the system operate.
kammer777
7 Posts
0
November 30th, 2011 06:00
This is definitely a limitation. With the virtualized nature of the entire system it should not be too difficult to give us a checkbox that would allow to turn on "short stroking" in mixed disk capacity tiers.
And, the problem will only become more pronounced as more of the lower capacity drives become unavailable.
avitarel
1 Message
0
December 12th, 2011 08:00
From a business / financial viewpoint, it would be a shame to purchase 450 GB disk and only utilize 300 GB. The 300 GB FC drives have not been available since June 2010 and will be no longer be supported by May 2015. How long have the FC disks been in service? Are they fully depreciated? At some point you will have to phase out the older FC disk. Since you have already started adding SAS disk, now may be time to build on the SAS disk configuration and start phasing the FC disks out of the system. This means you may be managing two pools of storage for the time being but at some point the FC disks will have to go away.
You already have 450 GB SAS Disk installed - are they configured in the same pool? Are you using Data Progression?
StrandAssociate
3 Posts
0
January 13th, 2012 09:00
Isn't Fast Track basically stroking? Maybe you should try to find info on how Fast Track would work with disks of different sizes in the same storage pool? Also, are we sure that the larger disks aren't automatically short stroked? I'm not sure if they are or not, but it's my understanding the Compellent array can specify where on the disk it wants to write a block, so wouldn't it make sense to automatically put it towards the center of the disk first? If the data is put to the center first and the utilization is capped at 300 GB that would be the same as short stroking.