Visualizing Linux Performance Data In New Ways
One of the items I've been working on recently for Phoronix Test Suite 3.4-Lillesand is new ways to visualize performance result data generated by the many test profiles and suites available via OpenBenchmarking.org. Here's one of the new ways that was committed over the weekend to the Lillesand Git code-base.
With the 120+ test profiles and 50+ test suites available to Phoronix Test Suite users by default, generating lots of data isn't an issue. There's benchmarks constantly being carried out by the Phoronix Test Suite either publicly and then uploaded to OpenBenchmarking.org or by many parties behind their firewalls.
The problem in generating so much performance data is then analyzing and making use of the results. If you have a result file with easily dozens of different data sets, it can be a bit difficult. Besides showing each result individually within the Phoronix Test Suite Results Viewer or on OpenBenchmarking.org (e.g. a normal result file), there's been a few ways to provide an overview. For example, at the top of each result file with multiple tests, there's a simple table to show the individual data points and some indication of the "losers and winners" via color-coding.
There's also mini "overview" graphs of each numerical result.
And also various analytical options from the Phoronix Test Suite client or the OpenBenchmarking.org management panel, etc. What's new to Lillesand now though are basically modified radar/spider charts that are automatically generated where relevant.
If you're running a lot of different tests with multiple systems/configurations, the Phoronix Test Suite will automatically generate one of these new charts for you within the results viewer. It provides a simple and effective means of spotting any clear trends or for quickly narrowing down to a particular area of interest. The Phoronix Test Suite will automatically group the results of the same test sub-system together and then within there pairing similar test results. All of the quantitative data is then normalized and presented in a clear higher-is-better format with the performance being relative to the slowest system in that workload. With the SVG back-end, mousing over a particular bar will indicate the specifics for that test.
It's very easy to spot the best and worst performers. The most relevant hardware details are also automatically rendered as part of the chart (e.g. if you're running a lot of CPU-centric tests, the CPU for each system will be automatically shown on this convenient display).
Various other tweaks and improvements will follow in the coming days. This was actually a very easy addition to the Phoronix Test Suite thanks to the pts_Graph framework and then the multi-format bilde_renderer image rendering library, also developed as part of the Phoronix Test Suite.
There's also a few other means of visualization that I'll be working on and hope to have fully implemented prior to the official release of Phoronix Test Suite 3.4-Lillesand in September. Your feedback is also always welcome on ways of representing test result data or for other Phoronix Test Suite features. For those interested, you can fully take advantage of this latest feature already if pulling the latest Lillesand development code from the Phorogit repository.
With the 120+ test profiles and 50+ test suites available to Phoronix Test Suite users by default, generating lots of data isn't an issue. There's benchmarks constantly being carried out by the Phoronix Test Suite either publicly and then uploaded to OpenBenchmarking.org or by many parties behind their firewalls.
The problem in generating so much performance data is then analyzing and making use of the results. If you have a result file with easily dozens of different data sets, it can be a bit difficult. Besides showing each result individually within the Phoronix Test Suite Results Viewer or on OpenBenchmarking.org (e.g. a normal result file), there's been a few ways to provide an overview. For example, at the top of each result file with multiple tests, there's a simple table to show the individual data points and some indication of the "losers and winners" via color-coding.
There's also mini "overview" graphs of each numerical result.
And also various analytical options from the Phoronix Test Suite client or the OpenBenchmarking.org management panel, etc. What's new to Lillesand now though are basically modified radar/spider charts that are automatically generated where relevant.
If you're running a lot of different tests with multiple systems/configurations, the Phoronix Test Suite will automatically generate one of these new charts for you within the results viewer. It provides a simple and effective means of spotting any clear trends or for quickly narrowing down to a particular area of interest. The Phoronix Test Suite will automatically group the results of the same test sub-system together and then within there pairing similar test results. All of the quantitative data is then normalized and presented in a clear higher-is-better format with the performance being relative to the slowest system in that workload. With the SVG back-end, mousing over a particular bar will indicate the specifics for that test.
It's very easy to spot the best and worst performers. The most relevant hardware details are also automatically rendered as part of the chart (e.g. if you're running a lot of CPU-centric tests, the CPU for each system will be automatically shown on this convenient display).
Various other tweaks and improvements will follow in the coming days. This was actually a very easy addition to the Phoronix Test Suite thanks to the pts_Graph framework and then the multi-format bilde_renderer image rendering library, also developed as part of the Phoronix Test Suite.
There's also a few other means of visualization that I'll be working on and hope to have fully implemented prior to the official release of Phoronix Test Suite 3.4-Lillesand in September. Your feedback is also always welcome on ways of representing test result data or for other Phoronix Test Suite features. For those interested, you can fully take advantage of this latest feature already if pulling the latest Lillesand development code from the Phorogit repository.
20 Comments