The Ups and Downs of Trends - There are Soft and Hard trend limits

A trend of CPU utilization stops at 100%, right?  Well, yes, but it doesn’t stop in the same way that a trend of disk space utilization stops. 

As the average utilization of a server approaches 100% , transactions submitted to that server take longer and longer to execute, but they are unlikely to actually fail (unless specifically forced to do so by a timeout mechanism).  The same applies to a variety of other limits such as I/O rate, network traffic, memory utilization and so on.  Certainly the server cannot run at more than 100% utilization, but the world will not suddenly end when that limit is reached.  Indeed, mainframes supporting large mixed workloads are designed to run at 100% processor utilization for long periods.

On the other hand, if an application needs to write data into a disk or filesystem that doesn’t have enough free space to hold it, then typically the application will fail with a hard error, and some kind of recovery action will be necessary.  Depending on the sophistication of the application and/or the file system, this recovery may take place automatically behind the scenes, but it still may result in a variety of complicated administrative problems.

For this reason, a trend showing that a filesystem is predicted to be filling up may need to be dealt with more urgently than a trend showing that CPU utilization is approaching 100%. 

Lesson 3 – not all trends are equal.  Some values of 100% are more painful than others.

Look out for my future blogs where I’ll be tackling some of the more complicated features of trends like statistical measures of confidence in the trend.

I’ll also be looking at cases where trending just isn’t going to work, so that alternative approaches, such as modeling, are required.

Until then take a look at Athene, our capacity management and modeling software http://www.metron-athene.com/


Andy Mardo
Product Manager