Ask the Expert: Performance Calculations on Clariion/VNX

If you run into performance problems on your front-end ports (e.g. high bandwidth, QFULL errors), you might want to rebalance your hosts over the existing or new FE ports. We've recently performed this kind of a change for a customer together with Rob. Our blogs are up here:

Rebalance VNX storage paths without downtime | FastStorage

Migrating hosts to new storage ports, LIVE | 50mu - about storage and how to access it

Hi all, I Have a question regarding the throughput consideration,

Say there are 5 SSDs(3500 IOPS) with raid 5 , 5 SAS(180 IOPS) with raid 5 and 8 SATA(90 IOPS) with raid 6 configuration in FAST VP. Let there 50:50 read:write percentage.

Total front end IOPS for all three will be:

SSDs= 7000

SAS=360

SATA=205.7

[Formula used: frontend iops= (backend iops/(reads + {writes * write penalty})]

Total IOPS = 7565.7

Let the block size be 8k

Therefore the throughput= block size * IOPS = 59.1 MB/s

So even though front end port can support upto 1GB/s  the maximum throughput this storage can support only 59.1 MB/s with all 8k blocks

similarly for 16k blocks throughput =118.2MB/s and for 64k = 472.8 MB/s.

My question is , Is this the right way to analyse storage throughput in terms of storage backend?

Hi Aneesh,

Your principe of calculating max bandwidth if you know the amount of IOps and the size of the IO is sound. It is indeed IOps * IO size = Bandwidth.

Also,  you're right if you're talking about a sustained, random I/O, no-cache workload: that specific back-end will not be able to max out the front-end ports in terms of bandwidth.

As for the maximum throughput of the system: keep in mind that the CLARiiON and VNX systems have cache. Cache has a much higher burst rate than regular drives. Also, not all workloads are a worst case example: sequential workloads are more friendly on the back-end and thus can deliver more data which can fill up your ports. Backup workloads are pretty much almost 100% read I/O which also allows for a higher number of IOps. So you might just be able to saturate those front-end ports

Hope this helps!

Hey Jon, thank you for the response. yes it helps ..

One more question i have is regarding the performance calculator that i have seen in few sites.

Say we take 5 SAS drives(180 IOPS) with raid 5.

100% read IOPS = 900

100% write IOPS = 225

50% read/write = 360

Formula used: frontend iops= (backend iops/(reads + {writes * write penalty})]

But when i did the same using performance calculator of a site my values matched with 100% read and 100% write but it didnt when it came to  50% read/write, it showed 562.5.

So after analyzing i came to a conclusion that they are coming to a linear approach i.e they are making 10 divisions of the difference of 100% read and 100% write and adding the value for for every difference in 10% read/ write..

ex: in this case 

100% read IOPS = 900

100% write IOPS = 225

their difference would be 675 iops

dividing the value by 10 = 67.5

so when i take 50% read/write what they do is multiply 67.5 with 5 = 337.5 and then subtract this value 900

i.e 900-337.5 = 562.5 !

So instead of considering the formula for each read and write percentage they have considered a linear approach.

Is there any reason for calculating it that way?

Hope i have made myself clear with what i had to ask!

Hi Aneesh, sorry for the massive delay, I've been on holiday for the last couple of weeks.

I think you've made an error in your calculation. If you would take 562 front end IOps at a 50/50 R/W, you'd end up with roughly 280 read IOps and 280 write IOps. Write IOps times 4 for the WP will already exceed the total IOps capacity of the RG.

Circling back to the 360 FE IOps in 50/50 R/W, you'd have 180 reads and 180 writes. 180*4 = 720; add the 180 reads and you're back on the total of 900 IOps. So this is indeed the correct value.

When I calculated I got 360. It is the calculators I found online that showed 562, which got me confusing!! Thanks a lot. It clears most my doubt. I even had discussion with one of the guys who had created the calculator, who said he was right. So came here to ask you guys , to make sure i wasn't wrong!!

Thanks a lot...

I was just stumped by this same question and I think you're right. Some sites with IOPS calculators use the formula below which doesn't seem right, and that's how they get the 562 in your example.

IOPS = (TOTAL_IOPS * %_Read) + ((TOTAL_IOPS * %_Write) / RAID_Penalty)

You are right, they're deviding by the write penalty, which is clearly wrong. You must multiply!

IOPS = (TOTAL_IOPS * %_Read) + (((TOTAL_IOPS * %_Write) * RAID_Penalty))

Hi...

I'm looking for a monitoring solution only for "VNX file OE performance" like data mover side ,  example: IOs, etc

I explored many tools in OpenSource like nagios etc, but didn't work as needed.

Please do suggest some options to have a performance monitoring for a group of 10 VNX File Arrays.

thanks,

UniSphere Central or VNX Monitoring and reporting

Thank you, Rainer.  Thank you so much.

Will Unisphere Central tool can be installed and used for a single point of managment console for around 40 VNX File Arrays?

Regarding the tool "VNX Monitoring and reporting" where can i find the information and installables.

Thanks for your inputs.

You can set up VNX monitoring and reporting under a trial license,

This application works with CX4...

Very good tools and easier to use than Analýser.

Another item you can check to get performance related data is the SAN ports! If you're using Brocade as your SAN, run the free tool "SAN Health" during the backup and make sure you have the tool gather performance data during the backup hours. For Cisco you might need Fabric Manager (or DCNM these days) to get performance data.

Hello

In VNX what is maximum utilization of single SP like ( SP- A or SP - B)  without any issue