[2810] in linux-scsi channel archive
Re: BT-958 or Flashpoint?
daemon@ATHENA.MIT.EDU (Kristian Soerensen)
Thu Nov 13 21:10:04 1997
Date: Fri, 14 Nov 1997 01:03:37 +0100 (MET)
From: Kristian Soerensen <elof@image.dk>
To: "Leonard N. Zubkoff" <lnz@dandelion.com>
cc: linux-scsi@vger.rutgers.edu
In-Reply-To: <199711132031.MAA29922@dandelion.com>
On Thu, 13 Nov 1997, Leonard N. Zubkoff wrote:
> Date: Mon, 10 Nov 1997 05:33:12 +0100 (MET)
> From: Kristian Soerensen <elof@image.dk>
[cut]
> Both 958 (Multimaster) and FlashPoint is supported. The difference is that
> the FlashPoint makes the PC CPU do some of it's work.
> This makes the card a little cheaper, but uses some CPU power when in use.
>
> If that's a worthwhile tradeoff is up to you.
> It might be a good idea with many windoze machines that doesn't do that
> much work while the disks is being accessed. Many people is eg. only using
> one program at a time in office work.
[cut]
> When the FlashPoint boards were first designed and released, it was the era of
> 100MHz - 133MHz Pentium processors and it was indeed true that the MultiMaster
> boards performed better. As processors increased in performance, the tradeoff
> changed such that the FlashPoint boards actually perform better in most
> environments. Inf act, for Linux the tradeoff was earlier since it is so
> efficient. As long as interrupt latencies don't grow too large, the FlashPoint
> boards generally perform slightly better in overall elapsed time for a task.
> On a heavily loaded server, the MultiMaster will probably still perform better.
> The MultiMaster boards are also slightly more robust in that having a
> completely separate processor makes error recovery more certain.
That's interesting. I must admit that I haven't tried a flashpoint in a
really quick PC, I didn't imagine that a contemporary Intel CPU
was enough to eat up the work-load difference.
Am I right in guessing that most dual- and quadro-CPU machines are better
served by a FlashPoint ? Unless the (rare) cases where their apps uses all
available CPU-cycles.
--------------------------------------------------------------
The Linux Resource Exchange http://www.falconweb.com/~linuxrx
Kristian Elof Soerensen elof@image.dk 45 93 92 02
