Understanding Performance VNX5200

Also is there a way to look at how much fast cache capacity is being used up? 

i see you changed your original reply, no need to get worked up ..just trying to help.

Yes I know that, I am not trying to look at fast cache on LUN level. 

Analyzer will not report Fast Cache hits for individual LUNs, you can look at storage pool FAST Cache hits only.

FAST Cache Dirty Pages on VNX2 uses a different FAST Cache method than the old VNX arrays. On the VNX MCx the FAST Cache acts more like a Write-Through cache than a Write-back cache. When you write to data contained in the FAST Cache, that changes the state of the data from "Clean" to "Dirty". Instead of keeping the "Dirty" data in FAST cache, the data is flushed to the disks very quickly, which makes the data in FAST Cache change from Dirty to Clear, but the data is still in FAST Cache waiting for more hits. You would see Dirty Pages spike up then disappear. This is the normal operation. This is also explained in the "VNX-MCx-Multicore-Everything.pdf" on page 29 in the "Page Management" section. You can find this document on emc.support.com.

https://support.emc.com/docu48786_White-Paper:-VNX-MCx-Multicore-Everything.pdf

FAST Cache may also not be caching the IO as it's too random - the IO needs to fall into the range where it will be promoted to FAST Cache. If FAST Cache is not all the useful - check the Pool FAST Cache Read/Write cache Hit Ratio to see if FAST Cache is being used - then you may be better off with the EFD disks in the Pool as the Extreme Performance Tier.

If you look at the disks in the Pool - separately, the 10K SAS can handle about 120 IOPS based on a 4KB IO Size and 80/20 Reads/Write random IO. If the IO size if larger or the IO is more Writes than reads and more random, then the amount of IOPS per disk will be lower. Response times will be higher as the IO Size increases or the percent of Writes is higher or if you have very lager Reads. There are a lot of parameters that can make the 10K SAS disks have higher response times. You should also look at the Queue Length and Average Busy Queue Length at the same time - if the ABQL is quite a bit higher than Queue Length, that indicates the IO is coming in bursts and that can also impact performance.

glen

Performance troubleshooting would require a lot of attention to your configuration. It's not just the drive's RPM that matters. From the RAID Type to to the number of drives and who owns what and everything should be checked.

If you need in depth analysis then it would need us capture the data and check with support. See what they find and what the recomedaons are. Test and see how it goes.

Hope this helps.

Regards,

Sheron

Hey Brett,

Does it take NAZ files as well?

Regards,

Sheron

I got the answer, it was up there in the page !!!D'oh !!

I've always use Analyzer and never tried this, It looks groovy n worth a look.

Thankyou for the link

There's no real way to see how much cache is used/available.

If you want to get a good overall picture of the performance profile of your array, without spending hours with Analyzer, consider giving Mitrend a go.

You'll get workload details, LUN details, SP details, Cache and FAST Cache details and a myriad of other counters and various heat maps. All you have to do is submit some NAR files and SP Collects and the work is done for you... for free. And you get pretty graphs

For a sample and some instructions, have a look: Mitrend- Self Service Performance Analysis for EMC VNX/VMAX/Clariion | Pragmatic IO

Sheron wrote: I got the answer, it was up there in the page !!!D'oh !! I've always use Analyzer and never tried this, It looks groovy n worth a look. Thankyou for the link

No probs, for regular Customers, I think Mitrend is gold.

Awesome this is perfect I have monitoring and reporting which is about 10X better than analyzer, I hate using Analyzer, thanks for the mitrend link this is great.