SunshineM
2 Iron

CIFS IOmeter results

Jump to solution

Hello,

Can anyone please explain me exactly what the IOmeter results mean or how to interpret?

I have attached screenshot of the results that I have ran on;

1) CIFS File system

2) Existing on RAID6 NL-SAS

Created a File Share on the above File System > Mapped it to Windows machine from where I ran IOmeter

Ran against 512 kb with 70 % Read & 30 % Write ratio.

Difficulty is how shall I interpret results that IOmeter is showing;

Is it like below?

1) IOPs: 183.23 = 187.82 * (total disks in RAID 6 where my CIFS share exist) = 187.82 * 8 = ~ 1500 IOPs

So is my current RAID 6 capable of handling ~ 1500 IOPs ?

2) MBpS: 96.07

So is my current RAID 6 CIFS Pool capable of handling ~ 96 MBpS of traffic?

3) Latency: 5.45 ms

So my current RAID 6 can handle ~ 5.45 ms of latency, not below that?

I want to know by this test, taking into consideration above File size and read/write ratio;

1) How much max IOPs my CIFS share can handle ?

2) What is the worst latency ?

3) Maximum Bandwidth CIFS share can handle ?

Please help.

Thanks.

Tags (3)
0 Kudos
1 Solution

Accepted Solutions
sluetze
2 Iron

Re: CIFS IOmeter results

Jump to solution

1) IOPs: 183.23 = 187.82 * (total disks in RAID 6 where my CIFS share exist) = 187.82 * 8 = ~ 1500 IOPs

So is my current RAID 6 capable of handling ~ 1500 IOPs ?

--> no. This is already the number of IOPS your RAID6 could handle in this specific test. This may vary when you modify the blocksize in your test.

2) MBpS: 96.07

So is my current RAID 6 CIFS Pool capable of handling ~ 96 MBpS of traffic?

--> your RAID6 CIFS Pool in this Test-configuration was able to Handle 96MBpS. This can / could be limited by Network bandwidth available for your test. (you will get lower results with 100MBit/s network than 1GBit/s network or 10GBit/s network) this may also vary with the blocksize configured in the test.

3) Latency: 5.45 ms

So my current RAID 6 can handle ~ 5.45 ms of latency, not below that?

--> the average latency in this test was 5.45. Average means there were values above and below that.

I want to know by this test, taking into consideration above File size and read/write ratio;

1) How much max IOPs my CIFS share can handle ?

depends on the blocksize of your test.

2) What is the worst latency ?

-> see "maximum latency" in your results.

3) Maximum Bandwidth CIFS share can handle ?

--> you mean throughput and this has some dependencies.

before using iometer you should think about what your workload is and how it "talks" to your storage. use this values to define a test to get valid results.

0 Kudos
12 Replies
SunshineM
2 Iron

Re: CIFS IOmeter results

Jump to solution

Awaiting reply from anyone please.

Thanks

0 Kudos
sluetze
2 Iron

Re: CIFS IOmeter results

Jump to solution

1) IOPs: 183.23 = 187.82 * (total disks in RAID 6 where my CIFS share exist) = 187.82 * 8 = ~ 1500 IOPs

So is my current RAID 6 capable of handling ~ 1500 IOPs ?

--> no. This is already the number of IOPS your RAID6 could handle in this specific test. This may vary when you modify the blocksize in your test.

2) MBpS: 96.07

So is my current RAID 6 CIFS Pool capable of handling ~ 96 MBpS of traffic?

--> your RAID6 CIFS Pool in this Test-configuration was able to Handle 96MBpS. This can / could be limited by Network bandwidth available for your test. (you will get lower results with 100MBit/s network than 1GBit/s network or 10GBit/s network) this may also vary with the blocksize configured in the test.

3) Latency: 5.45 ms

So my current RAID 6 can handle ~ 5.45 ms of latency, not below that?

--> the average latency in this test was 5.45. Average means there were values above and below that.

I want to know by this test, taking into consideration above File size and read/write ratio;

1) How much max IOPs my CIFS share can handle ?

depends on the blocksize of your test.

2) What is the worst latency ?

-> see "maximum latency" in your results.

3) Maximum Bandwidth CIFS share can handle ?

--> you mean throughput and this has some dependencies.

before using iometer you should think about what your workload is and how it "talks" to your storage. use this values to define a test to get valid results.

0 Kudos
SunshineM
2 Iron

Re: CIFS IOmeter results

Jump to solution

Thanks very much. I am going in write direction to understand this.

Ok, blocksize I took was 512 KB with 70 - 30 % Read - Write Ratio. This is what you meant?

If yes, then I will try with different block size and read/write ratio.

Again want to confirm, that 183 IOPs is the maximum IOPs generated taking into consideration above block size, on the share?

thanks.

0 Kudos
sluetze
2 Iron

Re: CIFS IOmeter results

Jump to solution

183IOPS was the average value your test showed. no maximum. maximums are also very imprecise (in my opinion)

if you lower the Blocksize you will get higher IOPS. In your Test every IO you send has 512K of data. thus you may saturate your network which limits the amount of data and IOPS you can transfer. If you lower your blocksize you need more IOPS to transfer the same amount of data.

for a 1MB file at 512K you need 2 IOPS

for a 1MB file at 8K you need 128 IOPS

the blocksize depends on the application. Windows Explorer network copies normally use 64K Blocks (depending on some tcp and smb settings).

this is why I said look at your normal workload and try to find a testsetting which matches your workload.

a DNA-Sequenzer uses higher blocksizes (~streaming)

a Database normally uses lower blocksizes

if you have a smb only environment and only use your storage for homedrives and groupdrives I would use 64K as a size to determine my iops.

if you have a specific application take a look in the documentation if there are any values to blocksizes.

0 Kudos
sluetze
2 Iron

Re: CIFS IOmeter results

Jump to solution
Rainer_EMC
5 Rhenium

Re: CIFS IOmeter results

Jump to solution

Doing a benchmark for IOPS it doesnt make sense to use 512k block size since applications that IOPS bound or sensitive use much smaller block sizes

Similar for thruput – if your app wants highest thruput it uses large block sizes

Bottom line: anyone can run a benchmark – if you don’t know what your workload is and how to interpret its just numbers without meaning

0 Kudos
SunshineM
2 Iron

Re: CIFS IOmeter results

Jump to solution

Our application requirement (given by that app vendor) will have below;

Host IOPs: ~ 1000

Block Size: 512 kb

Read / Write ratio: 70 / 30

Latency required: below 2 ms

Throughput required: 60 - 80 MB/s

Based on this, I know how to calculate storage design requirements, like RAID group, no. of disks, LUNs etc.

But I want to know if I can use my current RAID 6 Storage Pool having NL-SAS disks to host above application or not?

So I used 512 kb block size with 70 / 30 read/write ratio and generated the results with IOmeter.

Real-time CIFS performance will only tell real time data and not approx maximum capability (which I need to know)

So for that reason, I want to know the approximate maximum capability in terms of IOPs, Throughput, and Latency, the share created on this RAID 6 pool can handle.

Please suggest.

Thanks.

0 Kudos
sluetze
2 Iron

Re: CIFS IOmeter results

Jump to solution

done the math:

1000 IOpS with 512K blocksize each is 500MB/s.

something is wrong with my calculator or with the app vendors requirements.

0 Kudos
bhowen
1 Copper

Re: CIFS IOmeter results

Jump to solution

is this a simple kb vs KB confusion?  maybe they are looking at a 64KB block size... that may fall into a 60-80 MBps.

0 Kudos