We are planning to deploy a system for a client. I was told that leaving unnecessary modules (ie. jfs, usb, etc...) will not cause latencies in the kernel however the client is very sensitive. They care about latencies in microseconds and I was wondering if leaving unnecessary modules loaded would cause it and how can I measure the kernel latency (with lmbench if possible) w/o necessary modules. TIA
Hi... On Mon, Jan 3, 2011 at 20:06, Mag Gam <magawake@gmail.com> wrote:
We are planning to deploy a system for a client. I was told that leaving unnecessary modules (ie. jfs, usb, etc...) will not cause latencies in the kernel however the client is very sensitive.
That is right...module is only loaded as needed..
They care about latencies in microseconds and I was wondering if leaving unnecessary modules loaded would cause it
if you just "leave" them, it wouldn't add any overhead IMHO. Only if it is loaded. For that, you could either blacklist them or better....deselect them during kernel config. Whichever suits you better PS: If you're really sensitive, perhaps you could make all of your modules build into the kernel image...therefore there won't be any modules. But beware, the kernel image will be damn big :)
and how can I measure the kernel latency (with lmbench if possible) w/o necessary modules.
what latency are we talking here? scheduling latency? -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com
I am talking about scheduling latency and more important network latency On Mon, Jan 3, 2011 at 3:06 PM, Mulyadi Santosa <mulyadi.santosa@gmail.com> wrote:
Hi...
On Mon, Jan 3, 2011 at 20:06, Mag Gam <magawake@gmail.com> wrote:
We are planning to deploy a system for a client. I was told that leaving unnecessary modules (ie. jfs, usb, etc...) will not cause latencies in the kernel however the client is very sensitive.
That is right...module is only loaded as needed..
They care about latencies in microseconds and I was wondering if leaving unnecessary modules loaded would cause it
if you just "leave" them, it wouldn't add any overhead IMHO. Only if it is loaded. For that, you could either blacklist them or better....deselect them during kernel config. Whichever suits you better
PS: If you're really sensitive, perhaps you could make all of your modules build into the kernel image...therefore there won't be any modules. But beware, the kernel image will be damn big :)
and how can I measure the kernel latency (with lmbench if possible) w/o necessary modules.
what latency are we talking here? scheduling latency?
-- regards,
Mulyadi Santosa Freelance Linux trainer and consultant
blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com
Hi.. On Tue, Jan 4, 2011 at 10:15, Mag Gam <magawake@gmail.com> wrote:
I am talking about scheduling latency and more important network latency
Please don't do top posting.... About measuring latency, I think you can use lmbench along with ftrace (check http://lxr.linux.no/linux+*/Documentation/trace/events.txt for the details). Uhm and I think you could probably use the new "perf" tool, which is bundled in vanilla kernel tarball. Network latency? Perhaps you can check this article "Benchmarking network performance with Network Pipemeter, LMbench, and nuttcp" (http://www.linux.com/archive/feature/144532) All in all, the things that might affect the latency stuffs IMHO are: all tracing related functionality, traffic filtering/shaping, security modules. even if they were indeed loaded, I think the overhead might be negligible if you use them carefully and properly. -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com
participants (2)
-
Mag Gam -
Mulyadi Santosa