3 ms·
Five Years Of Linux Kernel Benchmarks: 2.6.12 Through 2.6.37
- kevinelliott 16y agoIt's actually quite disappointing to see that the Linux kernel performance essentially never increased over 5 years. A lot of nice functionality was added, so at least the lack of a net loss is quite fantastic. I wonder if this report might give the kernel folks the boost they need to be truly innovative moving forward.
- roadnottaken 16y agoSimple way to get me to NOT read your article: give it a flash-based popup-ad and split the article into 8 separate pages. I just want to see the data. Fail.
- rw- 16y agophoronix.com sucks anyway :)
- dododo 16y agothe choice of some of these benchmarks is weird. they have a bunch of cpu-bound benchmarks. why exactly would you expect these to change as the kernel changed? (i would expect changing compiler to have a bigger influence.) the ones that do change (and improve mostly) are ones where the kernel actually plays a role--file system, network, etc. the iozone benchmark is a bit surprising, it shows no improvement. but they are hitting 200MB/s, so i wonder if this is simply because linux is hitting the max speed of the disk. all in all, not very useful.
- bconway 16y agothe choice of some of these benchmarks is weird. they have a bunch of cpu-bound benchmarks. why exactly would you expect these to change as the kernel changed? Scheduling is a non-trivial task, and the kernel scheduler underwent significant work during 2.4 and early 2.6. I think the reason those graphs are so boring is because the tests involve single process/single thread benchmarking.
- wmf 16y agoIn general Moronix doesn't know what they're doing, but such benchmarks can be useful as regression tests; if the kernel isn't giving you full hardware performance then you have a problem.
- tlb 16y agoPlease standardize on either lower=better or higher=better. The graphs switch conventions frequently making it hard to interpret at a glance.
- shawndumas 16y agoI wonder how OSX's and WinNT's would look?