[kwlug-disc] Syscall load?

[Mikalai Birukou]

>>> > What about syscall load?
>>> >
>>>
>>> Do you mean as reported by the time command?
>>> Or do you mean something else?
>>
>> I believe this would be reported by the strace command, tracking
>> read and write syscalls. My theory is that with a varying blocksizes,
>> it will impact the number of trips into kernel space, and therefore
>> CPU load.
>>
>> A fast machine may be able to keep up with, and saturate, available
>> disk bandwidth. So even with the same throughput, CPU load and syscall
>> counts may differ. If so, then a slower machine may care about blocksize.
>
> That makes perfect sense: the more round trips into the kernel and out
> the more CPU utilization.
>
> I just thought that "syscall load" is a metric that can be displayed by some
> utility, like load average, hence the question.

Is it going to kernel, or,
is it sync to device operation that kernel (I expect) does with every block.

It feels that this device ops, their scheduling, that influences overall latency.
As device operations are orders of magnitude slower than traversal of
userland-kernel boundary.
Or, is this feeling misleading?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kwlug.org/pipermail/kwlug-disc_kwlug.org/attachments/20250903/0113cfb2/attachment-0001.htm>