Performance testing of SOC machines

Question

I would like to know a way to measure performance of SOC machines. We are in process of moving from physical servrs to VM Servers.

Thanks Jay

Answer 1

You might take a look at the Building a GIS: System Architecture Design Strategies for Managers Excel worksheets and presentations - there is tons of info there on capacity planning.

Answer 2

Aside from that spreadsheet; I find just doing good solid tests are a better indication. I have a statewide dataset I work with; with good known areas that I can build against and can log and track the time to complete a given set of tasks.

Be it building a cache; batch geocoding or other geoprocesses.

The Capacity spreadsheet is good for planning in ballpark terms; but there are a lot of things hardware/network wise that only good hands on testing and analysis will show you. Such as storage speed; do you have 15K HDDs, or are you using a SAN with 7200's or a iSCSI system with 10ks.

For me in the testing I am doing for my work between physical hardware and VMs, I am just building VMs that mirror the spec of my physicals and throwing like tasks at them to ensure that they respond the same way; being hits to a rest endpoint to return tiles or soap requests to a webservice for Geocoding.

Answer 3

Here's another one that @D.E.Wright got me to thinking about. Mxdperfstat from the arcscripts site. At a former job, a buddy used this a lot to test the performance of rendering dynamic data from SDE vs. file geodatabases, symbology, labeling, tiling - all kinds of metrics. If I remember correctly it's a ASP web app that you deploy on the server. From the mxdperfstats ArcScripts page:

Summary

MXDPERFSTAT (ArcGIS 10PreRelease and 93) can help diagnose typical MXD document performance problems, e.g.

• Inefficient scale dependency
• Slow symbology
• Large features
• Projection on the fly
• Potential database tuning

System Requirements:

1. Microsoft .Net Framework
2. ESRI .NET Assembly 10 or 9.3.1
   • ArcGIS Desktop with .Net support or
   • ESRI Engine Runtime
3. ESRI license:
   • ArcGIS Engine runtime or
   • ArcGIS Desktop

Answer 4

We have had decent success using JMeter, which can generate a constant stream of REST requests for random extents in your map service. We used a workflow similar to the one described here: http://gispunt.wordpress.com/2009/09/22/testing-your-map-service-with-jmeter/

That blog post provides a pretty good sample project (the comments and variable names are in Dutch, but it's mostly straightforward and Google Translate got me through the trickier parts). You will probably want to adjust some of the constants for your specific services. It may also be a good idea to test a variety of services (cached vs. dynamic, more complex vs. simpler, vector vs. raster, etc.)

Let me know if you decide to look into this and have any questions -- JMeter can have a bit of a learning curve, but once you get the hang of it, it's a powerful tool.