I consider myself knowledgable about SVC and am a huge advocat for it. I love our SVC environment and know that SVC helps me manage my SAN environment ten fold, however, we never invested into TPC to really measure out disk usage. I hope so much we will soon, but the outlook is grim. I did see in the Redbooks that it says RAID10 and RAID5 are very similar as far as performance is concerned when behind an SVC. Is this true?
Also since everything goes through the SVC and the SVC is limited to as fast as the cache can run, is it possible the SVC could be a bottle neck for some faster disk?
When going through the SVC, could servers notice a performance difference between 10k disk vs. 15k disk? I did read in the redbooks where it says SVC could actually speed up SATA disk because of the cache algorithms.
I'll take a shot at your questions...If anybody want to add or chime in, feel free...
Background for some of the answers: You state an incorrect assumption: Cache is not the limiting
factor; fiber bandwidth is. Cache would only become a limiting factor when over-comitted.
Also, you don't say what disk subsystems are behind your SVC, but they have cache, too.
Q1. The SVC, because of its pairing and private UPSes, never waits for a write ack from local disk.
As soon as its in cache, the host goes on as if it is on disk. Only on reads, not in cache (either),
would the SVC wait for disk...
Q2. As I said in the Background above: fiber bandwidth and especially the moving disk are the
limiting factors...
Q3. Not really a question about speeding up SATA, but with double cache and read ahead into cache,
you can see how the speed-up occurs....
Thanks for the reply. You did clear things up on Cache...it is not the limiting factor, which then gets me to my ultimate question: With all the cache between SVC and all the storage controllers (our environment has DS4800, 4700, 8100, and EMC CX4-240) will the host/users even see a difference in disk type? That is really what i was getting at when i said "it will go as fast as the cache can go" As you said once the disk writes to cache, it is ack'd by the host. So in a way...Having 10k or 15k disk isn't going to matter to the host, but rather the SVC nodes, which then they will send to the cache of the controllers.
So what i am getting in conclusion is that for disk writes, it really doesn't matter the speed of your disk, but for reads it could matter because of access time.
You will never really know if the hosts will receive the performance you calculate with.
Here are some rules for the planing of max. performance of a disk system:
1. RAID 10 vs. RAID 5
For read operations both are nearly equal. For write iops RAID 10 is better. With RAID5 the subsystem has to read the hole stripe, recalculate it and write it again and RAID 10 is just updating one block.
2. 10K vs. 15K
150000 rpm are 50% more than 10000 rpm. And at the end all the data has to be written on the disks. Calculate with 100 IOps per Disk for 10K Drives and a about 200 Iops per Disk for 15K Drives. This is no exact value but its enought for a estimation.
Your statement is true with the caveat that it still might be in cache because of usage or read-ahead pre-fetch....
It's also not as clear cut as the other responder, RAID10 has to do 2 writes...
The rule of thumb is that at about 5 RAID5 drives and above, RAID5 has a better usable disk/performance rate than RAID10. Especially if the stripes are tuned to be in sync (case in point, Windows messes this up...) or the O/S has single or multiple stripe size of the directly atached RAID5 drive.
This is NOT the case for the SVC, but then it can also mix DS8000, EMC, DS4000, etc, for one disk....
We can talk a lot about Performance and Caches an rpm's but at the end you will have to get some performance data of you environment. SVC is a fantastic product, but it's as fast as the subsystems in the backend are. If you use SATA drives you will have the performance of SATA drives. The Cache of the SVC can help but at the first backup - when the cache will be flooded within seconds - you just can get what the disk drives are able to give.
So my recommendation is to use 15K Drives 146 GB and build a RAID5. With DS4000 use 9 or 5 Drives for each array if possible. The "best practice" is to use 4 or 8 data drives + 1 parity. Use 256KB block size for the Logical Drive and also change the cache block size to 32 KB. Create just one Logical Drive per Array.
In DS8000 also create just one LUN per Array. I did it the old way -> Array1 -> Rank1 -> ExtentPool1 -> Rank group/Server1 ->LUN1.
You will have lots of Extent pools, but behind the SVC it does not matter.
With this short guideline you will design a perfect storage system for a SVC.
Regards,
Djole
Tags
Use the search field to
find all types of content in My developerWorks with that tag.
Use the slider bar to see more or fewer tags.
Popular tags shows the top tags for this particular type of content or application that you're viewing.
My tags shows your tags for this particular type of content or application that
you're viewing.
