Which Is Faster: Debian Linux or FreeBSD?

[Tanel]

This is one of the most interesting pieces I've read here on Phoronix. Thanks Michael, looking forward to the 8.1-one.

[Michael]

This is one of the most interesting pieces I've read here on Phoronix. Thanks Michael, looking forward to the 8.1-one.

You can thank those readers that don't use AdBlock and then Hewlett-Packard (for running HP printer US welcome screen ads on Phoronix through end-of-month) for motivating me to do a larger comparison like this. Some other interesting tests may be out next week as well besides the 8.1 ZFS testing and ATA-CAM benchmarks.

[smoe]

You can thank those readers that don't use AdBlock and then Hewlett-Packard (for running HP printer US welcome screen ads on Phoronix through end-of-month) for motivating me to do a larger comparison like this. Some other interesting tests may be out next week as well besides the 8.1 ZFS testing and ATA-CAM benchmarks.

I also want to thank for your work. What would very much help the points you are making would be to mention the number of repeats per measurement in the caption and to show the standard deviation in the plot.

This is a bit beyond Excel but very much worth it. I can help with an R script if you send me a fraction of the raw data.

Best,

SM

[Michael]

I also want to thank for your work.

You're welcome.

What would very much help the points you are making would be to mention the number of repeats per measurement in the caption and to show the standard deviation in the plot.

This is a bit beyond Excel but very much worth it. I can help with an R script if you send me a fraction of the raw data.

Look at the Phoronix Test Suite test profiles for each test to see the number of times the test is run (usually at least 3 or 4 times), but if the standard deviation between test runs exceeds 1.5% during that time, the run count will be increased (up to 2x the default run count) to lower the standard deviation.

With the new PTS3/Phoronix Global the standard deviation numbers as well as the run count and individualized run values will be much more accessible and easy for you to look at -- assuming you are in a web-browser that is very SVG capable A few of the basics can already be seen in Phoromatic Trackers with different mouse-over information like at http://kernel-tracker.phoromatic.com/

[Shining Arcanine]

This is the first article that I have read on Phoronix in a long time where Phoronix actually did a good job in terms of doing useful benchmarks. I planned to post a very nice comment here as I was reading it until I read the conclusion.

The benchmarks make it clear that a GNU userland is faster than a *BSD userland, which means that the *BSD userland developers need to improve performance. It also shows that the Linux and FreeBSD kernels are roughly equivalent in terms of performance, with Linux being better in file system performance and FreeBSD being better in other areas. Why is it that despite the clarity of these things, Phoronix makes an ambiguous conclusion?

[waucka]

You can thank those readers that don't use AdBlock

Hey! What about us subscribers?

[Michael]

Hey! What about us subscribers?

I'm always most grateful for the premium subscribers!

[kraftman]

with Linux being better in file system performance and FreeBSD being better in other areas. Why is it that despite the clarity of these things, Phoronix makes an ambiguous conclusion?

The file system matters when comes to file compresion and Freebsd was sometimes faster there. I suppose Linux with Ext4 will beat Freebsd there, but then, it will loose in SQLite. I don't agree with your conclusions. Linux was much better in areas not related to file systems.

[Shining Arcanine]

The file system matters when comes to file compresion and Freebsd was sometimes faster there. I suppose Linux with Ext4 will beat Freebsd there, but then, it will loose in SQLite. I don't agree with your conclusions. Linux was much better in areas not related to file systems.

I am not sure what benchmarks you read, but the ones I read had Debian kFreeBSD consistently outperformed FreeBSD in all compression benchmarks. Perhaps you are confusing the charts where lower is better with the charts where higher is better.

I was particularly interested in the GNU userland versus *BSD userland aspect of the benchmarks, because that is where there is a night and day difference. If you want to make this into a Linux versus FreeBSD comparison, the only places where Linux was both significantly and consistently ahead were the SQLite, Random I/O and dcraw benchmarks. Other areas are not quite so conclusive, because it could be well ahead on one platform, but identical on another.

By the way, here is an interesting question. Are the differences in the *BSD and GNU userland performance because of compiler optimization differences or differences in actual code?

[aragon]

The benchmarks make it clear that a GNU userland is faster than a *BSD userland

How so? I'm still seeing GCC 4.2 vs GCC 4.4 being a factor worth consideration in this benchmark.