Re: "real time" patches & naughty drivers

From: Michael T. Babcock (mikebabcock@fibrespeed.net)
Date: Tue Jul 18 2000 - 10:36:44 EST


I suspect that most people who use UDMA don't realise that it requires
as much CPU usage as it does.

That said, doing a "time hdparm -t /dev/sda" on my RAID 5 set gives:

23.44MB/sec

0.010u 0.680s 0:06.88 10.0% 0+0k 0+0io 414pf+0w

... under 0.7 seconds of system time during a ~ 7 second test at 10% cpu
usage.

That's what speed means. I'm not talking about straight throughput ...
but if you use software raid (which was the question; "why use hardware
raid?"), you're going to have much higher CPU use.

(Now to make it Kernel related:)

That and high-end SCSI cards support protocol sets that are great for
multi-threaded systems like command queuing and elevator sorting, etc.
These are features creeping into the kernel code, but I'd much rather
have them in hardware and keep my kernel running at near 0% CPU time
handling memory and security alone, letting my E-mail, web and database
server daemons have as much CPU bandwidth as possible to themselves.
Having a kernel that doesn't do these things eliminates a big piece of
latency in read requests, etc.

I'd love to have the kernel allow me to shut off "intelligent" functions
that are being handled by my hardware, but this has been brought up
before and I've been put down by people using (as is their right) less
powerful hardware who don't see the usefulness. And yes, I know I could
sit down and hack it in myself. But I don't get time in the week for
that, so I'm relatively happy with what I've got ...

Just my not-so-humble opinion.

Mark Hahn wrote:

> so what speed do you see?
>
> > CPU usage is almost nil on a Raid 5 setup ...
>
> OK. raid5, that is robustness, is a good reason. I suspect most
> people who do raid, even who do scsi, don't realize that a single
> udma disk can sustain 35 MB/s today. I'm guessing your HW raid
> doesn't come close to that, but I'm just guessing...
>
> > > out of curiosity, why are you using HW raid?

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Sun Jul 23 2000 - 21:00:11 EST