This post is more than 5 years old
9 Posts
0
2704
FE director utilization is high till 66% for 4 adaptors.
FE director utilization is going high till 65% for 4 adaptors. When I drill down each port on these frontend directors, Port 0 has 174 devices and Port 1 has 17 devices listed. Is the high utilization because of this imbalance, if yes how can reduce it.
PedalHarder
465 Posts
0
January 4th, 2016 19:00
Several factors can influence the %CPU busy on the FA:
- Read /Write ratio. Writes are more expensive than reads, so maybe the 4 FA's have a higher write rate?
- Array management. Does your Solutions Enabler host(s) use the 4 FA busy FA directors? Commands to the array are processed by the FA's and add to CPU overhead
- Open Replicator migrations
- VAAI operations (SCSI T10 type operations offloaded to the array like extended copy). ESX makes good use of VAAI operations for example.
How far you push the FA's before a performance management intervention is somewhat dependent on how you choose to manage your environment (and how your customers complain about poor performance
). Typically, 70% is considered to be close to the 'knee of the response time curve', where I/O response time starts to increase exponentially. You can obtain / monitor array based FA and device response time metrics to determine if the FA's are too busy. Another excellent metric to review is the FA queuing metrics. Busy FA's mean IO's may need to queue before being executed.
If you have FA's that are under-utilised, it may well be worth considering moving some workload to avoid the performance problem call.
PedalHarder
465 Posts
0
January 4th, 2016 14:00
You didn't mention the array model... For VMAX, VMAX 2, port 0 and 1 share the same CPU. The combined workload port 0 and 1 produce the CPU busy. The imbalance in number of devices between port 0 and 1 is not an issue.
Just to confirm, you are looking at the "%Busy" metric on the FE director in Unisphere right?
The %busy is related to IOPS typically. Drill down to the devices and find the devices with the highest IOPS to help understand who is using the FE CPU.
If the director IO rate is high, then to reduce the %busy, you will need to include more FA CPU in the port group(s) or move workload to other port groups.
raman_kaur
9 Posts
0
January 4th, 2016 14:00
Thanks for the reply Jasonc. Its a VMAX 20K model.
I am checking these stats in ViPR SRM, VMAX ; Performance Dashboard.
Yes, its the Busy (%) for these ports.
The throghput (IOPS) seems to be almost equal to other FA directors. (approx 2400 to 2700 IOPS)
Director IORate is maximum at 3903.
so, do we need to migrate the devices to other FA's
raman_kaur
9 Posts
0
January 5th, 2016 14:00
Thanks for the information. We tried to migrate the workload from the busy FA's to others with lesser workload. They look balanced now.
I think the array management or Open Replicator was not the issue.
I drilled down to see the Read/write ratio and the FAs had 25k write IOPs and arounf 50K read IOPS.
We have majorly ESX environment so, VAAI might also be the cause.