Why You Need Cross Domain Analysis
Thursday, April 24th, 2008One of the powerful aspects of Akorri’s BalancePoint product is that it works across the technology domains within the data center. You’ll hear us refer to our technology as being “cross-domain“. This means we do not just monitor a particular technology within the data center, such as servers, but we monitor all “domains” including applications, servers, switches, and storage devices.
So why is being “cross-domain” important? The answer is pretty straight forward. Without looking at all the domains you can not get an accurate understanding of how your infrastructure is working as a complete system.
Let me show an example of this. One of the most popular i/o monitoring tools in the UNIX/Linux world is iostat. Server administrators use iostat to figure out how well utilized their storage devices are. The problem with this is that iostat was written for simple disk models, such as a Direct Attached Storage (DAS) device. Once we migrated to more complex storage models and introduced SANs, then iostat doesn’t produce accurate results anymore. If you used iostat to fix a problem or plan for future needs then you are probably in for a big surprise.
Figure 1 below shows a graph that was generated using the data from iostat. It compares queue depth vs busy time of the storage device. Notice all the blue points clustered on the far right. Analyzing this graph you’d assume that the drive is very busy and highly utilized. You might even make the assumption that this is the root cause of some of your performance problems and you may go out and buy more storage in order to offload the existing device. You’ll probably find out that it actually doesn’t solve your problem.
Figure 1
Figure 2 below shows another graph of the same test just from a different perspective. This time we are monitoring the traffic using a Fibre Channel Analyzer at the storage device itself. As you can see a very different picture. This time you see the storage device is about 80% busy, still highly utilized but probably ok.
Figure 2
So as you can see if you used your server centric tools to understand how your server was interacting with your storage subsystems you’d probably be making some very bad assumptions about it’s operation.
A lot of products these days are beginning to appear in the marketplace that claim they can help you manage your data center. Products like VKernal’s Capacity Bottleneck Analyzer have the problem of being domain centric. Capacity Bottleneck Analyzer can not possibly tell you with any accuracy how well performing or utilized your storage system is due to it’s server centric view.
In order to fully understand the operation of your complete system you need to have a true cross-domain view. My advice before you purchase any tool to help you manage your data center is to ask the question of the vendor “Does your product support a cross-domain view of my infrastructure.”
Subscribe to our RSS feed