[Top] [All Lists]

Re: Is sendfile all that sexy?

To: <kuznet@xxxxxxxxxxxxx>
Subject: Re: Is sendfile all that sexy?
From: jamal <hadi@xxxxxxxxxx>
Date: Sun, 14 Jan 2001 22:47:47 -0500 (EST)
Cc: <netdev@xxxxxxxxxxx>
In-reply-to: <200101142029.XAA25591@xxxxxxxxxxxxx>
Sender: owner-netdev@xxxxxxxxxxx

On Sun, 14 Jan 2001 kuznet@xxxxxxxxxxxxx wrote:

> Hello!
> > 256M.
> 8Mb file should be enough.
> At least, I do see real difference between the first and subsequent
> runs with 32MB file and 512M machine. You do not and this is funny.

Maybe it is because i keep sending the file over and over again for the
duration of the 15 secs.

> > I think the L2 cache on these machine is 512KB. So something slightly
> > larger (600KB)?
>  >2*L2 cache. Otherwise it is not invalidated.

I have 512K L2 and 16K L1 and i made the buffer size 2M. No difference
in terms of throughput; but CPU utilization went down to about 50%

> > How Large? 170M currently.
> Seems, your first experiment was 16M. It was reasonable.
> I have no idea why you did not see differences between runs,
> it is simply impossible.

I'll send you my hacked ttcp in another mail. Maybe i am screwing
something ...

> > tput: 68MB/sec (compare to 86MB/sec)
> Sorry? But what did you measure earlier? 8)

sorry, earlier means in the case where no ZC patch

> > So i would say that CPU utilization has improved incredibly, but
> > throughput has gone down.
> I have no idea, throughput grows here on similar hardware.
> And by the way, numbers are pretty similar, but inverted. 8)

Ok, let me mail you the code.


<Prev in Thread] Current Thread [Next in Thread>