| HPE Storage Users Group https://3parug.com/ |
|
| 7200 AO drive and CPG requirements https://3parug.com/viewtopic.php?f=18&t=2880 |
Page 1 of 1 |
| Author: | danletkeman [ Tue May 22, 2018 1:16 pm ] |
| Post subject: | 7200 AO drive and CPG requirements |
Hello, We have a 7200 currently running and I am looking at adding some disk and AO to our current configuration. We have 36x 450GB 10K FC and 54x 2TB 7.2K NL disks. All of our VM's are on the FC disks in 8 volumes and one CPG. All of our file storage is in two volumes and one CPG on the NL disks and we have 10 volumes in one CPG on the NL disks for our video surveillance storage. Currently everything is 90% full and the IOPs on the NL disks are maxed out. The FC disks are also very close to being maxed out. I would like to add 8 SSD's and 18 more NL disks. 1. From what I have read on AO, the configuration requires that all writes go to the FC disks. Do I need to initially move the volume to the FC disk in order for AO to decide what it will move to the NL disks? 2. If Our current setup is seeing 150 iops on the FC disks will I need to add more spindles in order to handle the same or more writes that will happen on the FC disks? Or do you think AO will solve this issue? 3. Would flash cache be a better option for my setup? Thanks, Dan. |
|
| Author: | MammaGutt [ Tue May 22, 2018 2:19 pm ] |
| Post subject: | Re: 7200 AO drive and CPG requirements |
1. No. It isn’t an requirement that all writes go to FC. It is recommended to use a FC CPG as the User/base CPG for AO volumes because that is usually where most arrays have most performance... As SSD prices have dropped and people have more SSD capacity some are even using SSDs as User/base CPG. Keep in mind that all new capacity allocations happen in the User/base CPG, but when a block(or region) has been moved up or down any overwrite of those data will happen on that tier. 2. AO will not reduce the backend IOPS. So if FC are running at 150 IOPS and NL at 75, AO will probably not do much magic. If you look are the Region density reports on CLI or SSMC you might see that you have some very active regions. If your drives are maxed out IOPS wise but still have free capacity, adding a SSD layer and AO might move the hottest regions to SSD reducing the load on FC (and NL) allowing you to fill those drives without a performance issue. The old golden rule a few years back was that SSDs provided the performance of 20x 15k drives with 10x the cost per GB.. that isn’t quite accurate anymore as 15k drives are almost gone and SSD prices are now maybe 2-3x the price per GB vs 10k drives. Only drawback is that the drives are larger (3.84TB vs 1.2TB) so the minimum SSD pool has a price point. 3. Flash cache is read-only. If most of your IOPS are read and a very small dataset, then maybe. I’ve yet to see a use case where AFC did a better job than AO but it does require less SSDs. Edit: any reason why you have so many CPGs? If you look at AO you should reduce it to prevent capacity being «stuck» in different CPGs and tiers, and reduce the number of compact CPG jobs you need to run to manage that. |
|
| Author: | danletkeman [ Tue May 22, 2018 4:09 pm ] |
| Post subject: | Re: 7200 AO drive and CPG requirements |
Thanks for the reply. It all makes sense. I do have a question on the ssd pricing. I will send you a PM. I meant to say volumes. I don't have that many CPG's. I will edit my original post so it make sense to anyone else reading this. |
|
| Author: | Proc_rqrd [ Thu May 24, 2018 1:42 pm ] |
| Post subject: | Re: 7200 AO drive and CPG requirements |
to add. the density of SSD is high now. it is possible to use the SSD option as your "VM" Tier and the FC/NL AO config across those or all tiers. it depends on how many free slots in how many cages you have. or are willing to add with ur expansion. Dedupe is also an option with SSD and ive had good success with our vm environment on ssd. AO and AFC do not overlap. so it is ok to use both and not just one. any ssd resident data will not go into fmp and eventually any warm/hot data will get picked up by AO and moved up to the ssd tier. so the random read behavior that AFC tries to mitigate is beneficial. the amount of benefit does diminish as it scales though. so afc isnt a core fix for read performance. if you get ssd's it is really not that big of a deal to make a small portion of it AFC. if your array is running warm with a good read rate, any fmp cache hits are good. i have some with 15% or higher fmp's consistently. ill take it for what little storage i invested in the function |
|
| Author: | Loudermil [ Sat May 26, 2018 2:31 pm ] |
| Post subject: | Re: 7200 AO drive and CPG requirements |
danletkeman wrote: Thanks for timely male enhancement pills recommendations and the reply. It all makes sense. I do have a question on the ssd pricing. I will send you a PM. I meant to say volumes. I don't have that many CPG's. I will edit my original post so it make sense to anyone else reading this. How's the "wear and tear" of the SSD's when you do this btw? Or does it not play any role really? |
|
| Author: | MammaGutt [ Thu May 31, 2018 7:11 am ] |
| Post subject: | Re: 7200 AO drive and CPG requirements |
Wear on SSDs are generally because of writes. AFC should have lots of reads (as it is read-only cache) and not so much writes. |
|
| Author: | AlexFloyd [ Thu May 31, 2018 1:19 pm ] |
| Post subject: | Re: 7200 AO drive and CPG requirements |
danletkeman wrote: Hello, 3. Would flash cache be a better option for my setup? Thanks, Dan. AFC and AO is a complementary. On 7200 system AFC has a limitation 768GiB. So you can provision some space for AFC and other CPG/VV. Some VV can be tiered, other can use a AFC. In such case you should create separate CPG for each tier for AO configuration. But note that different type of IO (i.e/ IO block size) can interfere with one another. PS: Of course tiered VV can use AFC too. |
|
| Author: | AlexFloyd [ Thu May 31, 2018 1:37 pm ] |
| Post subject: | Re: 7200 AO drive and CPG requirements |
danletkeman wrote: Hello, 2. If Our current setup is seeing 150 iops on the FC disks will I need to add more spindles in order to handle the same or more writes that will happen on the FC disks? Or do you think AO will solve this issue? Thanks, Dan. Of course, additional SSD drives (AO ld's on SSD) will improve throughput, but in such heavy system the bottleneck can be on a cache layer or on a back-end layer Look on Delay ACK on a cache report. |
|
| Page 1 of 1 | All times are UTC - 5 hours |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|