I am working on a project adapting system software like the Linux kernel. I read one paper titled "Experience on Comparison of Operating Systems Scalability on the Multi-core Architecture" that tells how to detect bottlenecks using benchmarking. I am following their experimental approach and am able to generate a profiling report using the profile tool, but I am not able to detect bottlenecks from this profiling report.

Basically, I am not able to explore the no-vmlinux part, and I am not able to find those functions related to bottlenecks. Also I am not able to use lockstat.txt as there is no file in /proc/sys/kernel/lock_stat.


2 Answers 2


If you want lock stats on your kernel, you have to recompile with CONFIG_LOCK_STATS enabled. Look into Documentation/lockstat.txt in the kernel tree for more documentation.

  • Thanks for your support. I have ubuntu 12.04 installed on my machine. I want to use lockstat, I have looked at Documentation/lockstat.txt but I am not getting how to use it, if possible please help me out.
    – user33729
    Mar 7, 2013 at 9:21
  • Which part don't you get? Did you recompile your kernel? The option you need to enable is "Kernel hacking" -> "Lock usage statistics" in menuconfig/nconfig
    – vikraman
    Mar 7, 2013 at 13:17

Note that when the paper get published, some 3 years (average) has gone by since the original work was done. That is a long time for Linux development, don't take the results too seriously.

Look around, there are demented souls (I think at Silicon Graphics and similar outfits) (trying to) run Linux on 64K cores in the same machine (not a cluster!), I'm sure you'll find they flock together in some mailing list or IRC channel, perhaps they have some webpage. The "many, many cores" machines are of interest to people at IBM and intel, at least from what I've heard from near sources. It is also one of the interests of the regular kernel crowd, our normal PCs probably won't grow in clock speed but number of cores, and the two niches in which Linux has firm footholds are embedded systems and higher-end servers, so this is certainly a central issue.

You must log in to answer this question.