Blogs

Why does Infrastructure Performance Management matter?

You can’t measure performance without the right tools

By Ryan Perkowski, Senior Product Manager

Person 1 – “How was the commute this morning?”

Person 2 – “Oh, about 1000 cars per hour.”

What a strange way to answer this common question. Person 1 wanted to know how long it took Person 2 to commute to work. It was a question used to gauge how good or bad the traffic situation was for Person 2. We take this for granted in our everyday conversations, an implied request for specific information. How long did it take you to drive to work this morning?

What is strange is how Person 2 answers the question. “Oh, about 1000 cars” is not answering how long it took him to commute, it was answering how many cars made the commute. The statement ‘about 1000 cars’ doesn’t describe how fast or slow those cars were traveling, or how long it took them to traverse the drive into work.  It is merely a statement of how many cars made the drive. In this context, this is a VERY strange way of responding to ‘how was your commute?’

When an application in your datacenter is running slow, the application owner wants to find the source of that slowness.  When he/she asks the storage manager if the “storage system” is running slow, he is asking for a measurement of performance.  Traditionally, the only metric available to the storage manager is how many MB/s are flowing through his network. This is answering the commuting question with how many cars per hour.

Business-critical applications require all layers of the application stack (from database, server, network, to disk access) to be monitored for performance, not just throughput or errors.  Any monitoring tool worth its salt should be able to tell you if a hard error occurred within their space.  It takes an application-centric Infrastructure Performance Management (IPM) solution to tell you how well that space is performing.

Rather than answer the application owner with meaningless statistics, the storage manager, or network manager, or database manager should be responding with performance metrics.  Specifically, three facets of performance should be always measured.

  • How fast are the input/outputs (I/Os) being satisfied? In the storage world, how long does it take to complete a read or a write?
  • How large are these I/Os? In the storage example, this is a measurement of block or file sizes.  A larger block or file size will obviously take longer to complete than a smaller one.
  • How many I/Os per second (IOPS) are being completed? Is the system being flooded with requests, or is there only a trickle of data flowing (albeit good response times)?

Why do administrators answer performance questions with MB/s answers?  Simple, they lack the tools, and lack the visibility into what is occurring within their systems.  Infrastructure Performance Management looks to change all of that.  Your company has invested millions into its infrastructure.  It is time we start monitoring it properly.