how to do performance tuning in dmx?

Very general performance question

So to answer it generally, there are not a lot of knobs to tweak in Symmetrix. This is a good thing from a manageability point of view. The main ways to impact performance on DMX is through the choice of configuration options. Disk type, raid type, disk quantities, DA pair counts, front-end counts, cache capacity, etc.

You can also use QoS features such as Optimizer, Cache Partitioning, Copy QoS, Priority Controls, etc to manage the configuration once it is in place.

If you want to be a bit more specific, maybe we can help with some more specific answers.

DMX performances in a nutshell

Nice list Sanju!

However I'd suggest a formal perf-analysis training straight from EMC Education.
You gave good questions .. but finding good answers may be tricky. That's what our Perf-Analysis workshop does.

Before you go for performance tunning in dmx,one should anaylze the dmx performance.

so here are the some point which you consider while analyzing dmx performance.

Check Cache Performance:

Run the System à System Cache Utilization ViewGraph. You will see

What is the average cache hit percentage for this workload?

What is the worst cache hit ratio observed in this data?

Was the hit ratio inflated by write activity?



Check Balance Across the Front End:

Run the Dir-Fibre>All Fibre Directors-IO Rates ViewGraph. You will see

Is the activity constant over time?

Is the activity balanced across all the channels?



Check Balance Across the Back End:

Run the Dir-DA>All Disk Directors > IOs per Sec ViewGraph. You will see Is the activity balanced across the back end?



Check Balance Between Front End and Back End:

Run the System>All Directors ¿ IOs per Sec ViewGraph.

Is the front end more active than the back end?



Check Logical Volume Balance:

Run the Devices>Average IOs per Sec ViewGraph.

Are the average IOs per second spread uniformly?

Does there appear to be any significant striping?

What tool are you using to get this information?

must be performance manager (part of ECC)

You can also capture all those with the STP daemon which is now part of the solutions enabler kit.

It seems the daemon options are listed in the "Solutions Enabler Symmetrix CLI Command Reference" manual. Also see the man pages on your system. "man storstpd"

oh neat ..is there documentation how to do this in SE manual ?

I agree it is very generic question and a large topic. The general strategy laid out earlier is good, but generally you will need some threshold numbers to go with each of the components to figure out what you see is good or bad. If there was a reliable %utilization number that the metrics put out, it would have been easier to manage DMX performance. It is a good idea to figure out what the best threshold values are for your environment and then based on that monitor the performance profiles for various components. This is what I end up doing for each site and then create a utilization based heat chart that can give me an overall one page utilization view.

I will start with "It Depends", but here is a quick strategy.

The strategy that I use to do performance analysis on DMX is as follows (very briefly):

System Level
- Bandwidth - MB/sec should be well below the max recommended for the array. This is rarely an issue, unless connected to an high bandwidth requirement application going into 1-2GB/sec or more.
- Throughput IO/sec - In open systems env, if you start to go into tens of thousands of IOPS (30K, 40K etc) it should be looked into further. May point to bottlenecks in the front-end directors
- %Cache hit for open systems, anything below 70% will start to cause degregation, below 50% is bad.
- Any delayed deferred writes/sec is bad for performance

Front-End Directors (FA's, Ficons) - Similary create threshold values that work for you for IOPS and MBPS

Disk Directors (DA's) - Similary create threshold values that work for you for IOPS and MBPS

Remote Adapters (RA's) - Similary create threshold values that work for you for IOPS and MBPS

Disks
- Depending of RPM of the disks the IOPS per type of disk can be looked at to identify the hot disks. Use "SCSI commands/sec" to identify the busiest disk. For 10K RPM disks, I use a threshold of 120 IOPS for warnings
- Also look to see if most of the disks are active at all times to leverage all the spindles.

Devices
- WLA metrics show some service time metrics. it is a good idea to see if any of the devices showing high service time. This is highly dependent on application requirements but I mostly find anything in range of 5ms,10ms, 15ms, 20ms are okay in that order. Anything greater than >25ms should be looked at further.
- Write Pending Counts - Make sure they are below Max write pending count for the device. Upto 2-3 times should be looked at deeper

There are many more items to consider, but this can get you started. Bascally, creating a baseline of each of the array and then creating a baseline of each of the hosts is the best strategy. Then use these to compare against during a performance problem. I have created a lot of tools for Symmetrix arrays to easily manage multiple arrays and percolate up the issues with automated identification of hot components. More info at http://www.interscapetech.com, if you are interested.

Alok, welcome to the forums. When you get the change, be sure to introduce yourself in the coffee break area.

Will do, thanks for the welcome.

ARIFur, were the reply's to your post able to answer your questions? If so, please mark the helpful and correct posts and also mark the thread as answered. Doing so helps others identify what posts were the most useful.