About "Average Busy Queue Length"

Hi Santhosh Theyyan,

What is the good value of service time does the RAID Group have?

I will use following guide line to verify the storage performance:
1. All component's utilization(included SP, RG, LUN) < 70%
2. Drity Cache value are between 80% and 60%
3. All component's response time < 100ms
4. Total Bandwidth value of RG is < 12MB x 16(disk)
5. Total Throughput value of RG is < 180IOPS x 16(disk)
6. Disk Quene of physical disk < 10

Is above right?

Best Regards,
Dennis Dai

you can also check service time for the LUNs invloved using the navishere Analyzer

i am not that familier with those vaules from SAN side. Some experts would hopefully explain that..

But from the host side the service time of below 20ms is conisidered as good.

I guess you're using 15k spindels ? The 10k ones are around 130 IOps.
And you have 16 disk raid groups ? That's rather large, but supported. Best practice is RAID5 (4+1) and is ok in most cases.

Hi RRR,

Yes, the disk' RPM is 15k.

I like to propose the RAID 1/0 (8+8) to my client.
We will run db environment on that.

I have implemented some mirrors replicate from local CX3-80 to remote CX700.
One of hosts sometimes faced "100% I/O wait error", I am checking what's the root cause.

I guess the remote R1/0 (8+8) 73GB RPM 10K disks is the bottleneck.
And find out more evidence to prove.


Thanks for any help.
Dennis Dai

The numbers you quote are good, but I would lower the response times to say 20ms.

If you keep LUN/SP/Disk utilization to less than 70%, Queue Length to less than 10 IO's and response time under 20ms, that's our definition of Best Practice.

Make your Write cache as large as possible, and Read cache between 100MB to 250MB (lower is better as you have more for Write cache).

Most of this is found in the follwoing:

EMC CLARiiON Fibre Channel Storage Fundamentals - Technology Concepts and Business Considerations

http://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/H1049_emc_clariion_fibre_channel_storage_fundamentals_ldv.pdf

EMC CLARiiON Best Practices for Fibre Channel Storage: FLARE Release 26 Firmware Update - Best Practices Planning

http://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/H2358_clariion_best_prac_fibre_chnl_wp_ldv.pdf

regards,

glen kelley

When you mirror to the second array, you need to have the same speed disks - you need to take into account the latency of the link between the two arrays and the speed of the mirror disks. Slower disks will start to back up the Write cache on the mirror array - it will have trouble keeping up with 15K disks when writing. Make sure the Write cache on the mirror array has the maximum memory possible.

regards,

glen kelley

Hi Kelleg,

Thanks for your reply first.

There are something confused me.
Both Dirty_Page% of local and remote storage are under 80%, so
the write cache are enough to service production I/O.
But I can see higher response time(> 100ms) via Navisphere Analyzer.

Is this situation possible?

Thanks,
Dennis Dai

There are conditions where if IOPS are low, say under 75 IOPS, and Queue Length is low, you can get a false response time - the calculation is weighted to be more accurate with high IOPS.

You need to look at when you get high response time, what is the IOPS and Queue Length at the same time - if thses are low, then response time is probably incorrect.

glen

As far as I know there is limitiation on write cache size about 3000GB. So in CX3-80 if we allocate more than 250GB for read cache will improve performance or not?

CX3-80 max Write cache is 3072, you can assign the rest to Read cache. More read cache will will improve performance, but not as much as Write cache does.

glen

So of the 8GB per SP, 1 goes to the SP itself, 3 to write cache and 4 to read cache ?

On the CX3-80, yes - roughly

glen

I'm thinking out loud here.... why isn't it possible to dedicate let's say 6GB to write cache and 1GB to read cache ? What is the use of having 8GB if you can only have 3GB write cache ?