From owner-freebsd-hardware Thu May 27 19:15:40 1999 Delivered-To: freebsd-hardware@freebsd.org Received: from george.lbl.gov (george.lbl.gov [131.243.2.12]) by hub.freebsd.org (Postfix) with ESMTP id 384A715185 for ; Thu, 27 May 1999 19:15:37 -0700 (PDT) (envelope-from jin@george.lbl.gov) Received: (from jin@localhost) by george.lbl.gov (8.9.3/8.9.2) id TAA07899; Thu, 27 May 1999 19:15:28 -0700 (PDT) Date: Thu, 27 May 1999 19:15:28 -0700 (PDT) Message-Id: <199905280215.TAA07899@george.lbl.gov> From: jin@george.lbl.gov To: chuckr@picnic.mat.net, richard@pegasus.com Subject: Re: Support for Symbios vs. Adaptect SCSI Cc: hardware@FreeBSD.ORG Sender: owner-freebsd-hardware@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org On Thu, 27 May 1999, Chuck Robey wrote: > > } > In general, the PCI interface on the Adaptec chips is unsurpassed > > } > in the industry. > > } ... > > } > > } This is very helpful information. It does make it clear that Adaptec > > } makes sense at the high end. You get what you pay for, etc... > > > > I've seen test reports in the past that place the Symbios controllers > > at a slight speed *advantage* over Adaptec. (Sorry no references, too > > long ago.) > > > > Yes, the drivers could easily be the controlling factor. > > > > By most reports the Symbios controllers are very good, and not > > nearly as absurdly priced as Adaptec. > > > > Symbios chips can be found embeded in everything from workstations > > and pc's to large RAID units. > > Trouble is, your report has no specifics as to what is faster or better, > and even you don't recall where it's from, so it's of nearly no more use > than rumors. I'm not trying to be offensive, but (unless you have some > reason to assign more importance to your statement, that you haven't > passed on to us), this kind of stuff is misleading to folks. > > Those "test reports" could very easily have come from biased > benchmarking, something we're all familiar with. The fact that NCR > chips show up in many interfaces, well, they ARE cheaper, aren't they? > You have to have at least some kind of detail in reports such as yours, > to make the data believeable. The SCSI benchmark is very depended on the controller and drive combination. SCSI hard drive 1 may have #A throuthput on Adaptec controller, and #B throuthput on Symbios controller, but the SCSI drive 2 could have #B throuthput on Adaptec controller, and #A throuthput on Symbios controller. Unless someone tested a bounch of drive under both Adaptec and Symbios SCSi controllers, your never can tell the truth. One reason is that the disk dirve manufactories can make very different arbitrition timing as we found. Even you chained different SCSI drive on the same controller, you may end up having a serious trouble to make high SCSI bus utilization. Adaptec SCSI controller was very sensitive to the terminator and cable length a few years ago, so we stopped using it. We used Symbios instead. Since I have no curent Adaptec SCSI controller, I cannot tell how good it is. I only can tell we are happy with Symbios chips. We can get bits going almost saturating the SCSI bus at 85-92% utilization. I am very eager to hear if someone can tell us how Adaptec controller can do, so we may get one for benchmarking. So, at this point, unless you have done some intensive-combined SCSI benchmark, no one shall tell who is better. If you can get SCSI bus saturated, you are there. Another thing is the price/throughput ratio. In a low performance chain, you could chain up to 15 disks on a SCSI-3 controller. The bus is always saturated when all disks try to talk. The best throughput is 40MB. In a high performance chain, two high speed SCSI-3 drives can almost saturate a SCSI-3 UW controller. Now you need at least 7 controllers or 15 dsks; then the price is the big issue. The 2% differnt performence between different SCSI controllers can be ignored. In talking 1TB disk sub-system, I would not worry about the 2% or even 5% (could be seen at most) differnce of throughput for different SCSI controller. The price is more important. Just a coule of cents. -Jin at this point. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hardware" in the body of the message