DMX-3 Performance Monitoring

All DMX components should be monitored. It includes, fibre-channel ports, front-end directors, cache, back-end directors and physical spindles. As a general strategy, establish a baseline and monitor utilization growth and changes in I/O pattern over time. If a performance problem occurs, you can see what changed by comparing new I/O profile to the previous one.

By default, ControlCenter collects interval, daily, weekly and monthly data. Interval is the most detailed, every 15 minutes by default. Daily averages out every hour, weekly averages out across 7 days and monthly averages out across the entire month. For problem analysis, you would need to use interval data and for growth and changes in I/O profile over longer periods of time, you would use averaged-out data.

Symmetrix Performance Workshop is available for a more detailed discussion of performance monitoring and analysis.

This is a very loaded question and falls into the "it depends" category. Not aware of any best practices being available though the Performance course EMC offers is very good and does a good job of offering some good overviews.

As suggested already you can not really measure performance without setting a baseline and understanding your environment. What may be acceptable in some environments may not fly in your environment if your applications are very sensitive to response time.

Some good rules of thumbs:

1) Check shared components such as FAs and FA ports and make sure these are reasonably balanced.

2) Cache hit rate is a decent one to check and cache write misses are VERY good to catch as most applications will complain if they writing directly to disk. You can get this by checking Write Pending limits.

3) General view of drives to make sure you do not have any drives sitting for prolonged periods of time above the 70-80% utilization range.

Once again none of these alone are an indicator of a definite problem - if your host can tolerate high response times (a backup/archiving application) then any metric on its own can show.

Sorry for fairly generalized suggestions - but without knowing your environment or the types of applications it is hard to be more specific. I will say that there are some good whitepapers for tuning/monitoring specific applications (such as Oracle, etc.) available from both EMC and many other sources.

There is an EMC customer training class for Symmetrix Performance, you are better off attending that.

Hello, Thanks to all for your suggestions.