On 11/09/98 18:40:33 you wrote:
On Mon, 09 Nov 1998 13:25:57 PST, Sam Schorr sschorr@homestead.com wrote:
I think the reasoning here is not quite thorough enough. If I buy a Compaq server and then I install a 3-Com ethernet card, I still expect, and get, full support from Compaq. This has been the case in the clone, or Intel world for quite awhile. I know that it is less true the more proprietary you get, but still is essentially the case, even in the IBM mainframe world and in the UNIX world. I am now in the middle of an issue between Network Appliance and Microsoft and I would NEVER buy another filer if NetApp tried to deflect response to the fact that I use Microsoft products and therefore I should debug the Microsoft side first. Whether I like it or not, and whether NetApp likes it or not, multi-vendor environments MUST be supported
- there is no option.
I agree, in general.
I think netapp's position is somewhat justified, because they can claim -- and I take it on face value -- that they stress disks more than other vendors. Therefore, they claim, they have to test each model disk and even each firmware revision before quality assuring it. In theory, any generic disk should work, but there's enough "practice" out there to show that different disks stand up to various stresses differently.
I can accept this, but I'm not sure why this also applies to memory. And I guess I can accept it with scsi boards and ethernet boards, but I'm not entirely happy about it. It seems to me they could QA more add-ons than the ones they sell.
Actually, my impression (I have no data to back this up) is that the memory failures are roughly as likely as the disk failures, and they weed out a lot of memory in their testing as well. This is because they stress the memory as much as the disk (for the most part a lot of the memory acts as one big RAM disk), and memory timing issues need to be very, very tight.
SCSI and ethernet boards similarly have to be qualified. I agree more could be done, but there would be a tangible performance loss for running a 3com card in your filer instead. And Netapp would have to develop a driver for it and it's multiple incarnations. There isn't a nice driver interface layer you can use to make the driver standard across different platforms that you can also use for Netapp, because then you'd lose even more performance.
On a final note, you're not the only one I've seen doubt Netapp's "claim" of testing. I don't know why people doubt this. Trust me, I've seen it. You can probably call them up and ask to see it, too. There's a lot of QA that goes on.
Bruce