Re: Maximum number of syscalls per thread?
Re: Maximum number of syscalls per thread?
- Subject: Re: Maximum number of syscalls per thread?
- From: Michael Smith <email@hidden>
- Date: Thu, 10 Aug 2006 12:39:53 -0700
On Aug 10, 2006, at 12:02 PM, email@hidden wrote:
Just a quick question! I'm running an app which is used for testing
i/o. Right now, I'm running a thread at an elevated priority which
does a lot of pread() calls. The app works fine but is a little slow
& right now, I'm fine-tuning it. One issue, though; when I fire up
BigTop, find my process and look at the system call throughput, I
notice that it appears to be capped at 9000 calls/sec. I'm guessing
this is being throttled by a kernel parameter.
Your guess (with regards to a limit on the syscall rate) is incorrect.
I have observed a single-process syscall rate over 650,000/sec on
an intel Mini, so your 9,000 is barely ticking over.
The device I'm hitting is more than capable of
beating that, so I'm definitely not saturating the driver. Thoughts /
hints / ridicule welcome :-)
This sentence is logically invalid. The *device* may well be capable
of more than 9,000 IOTs/sec, but you haven't established anything at
all regarding the driver here. Have you actually examined the
interaction between driver and device to see what sort of utilisation
you're seeing?
Without more details of your test and driver architecture it's hard to
offer you more suggestions. You are aware that the clustering layer
and buffer cache will both affect the relationship between system
calls and I/Os? Have you exercised your driver to see what it's
maximal work rate is? Which system is your device in? How is
it connected? Whose driver, and how is that driver architected?
= Mike
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden