At 03:05 PM 9/27/02, you wrote:
>I don't care about a warranty.  That's great in a desktop, but in a server,
>a warranty means nothing.  If the server is down, the company isn't making
>money.  This varies depending on the company, but if they want *ME* to build
>them a server, then it's either going to be expendable (in which case I'd
>use a desktop), or it'll be a server.  Servers do not use IDE drives.  Heck,
>even Workstations move to SCSI.

The quality IDE is better than it use to be. Other than IBM Hungary plant. 
We haven't had much problems with either IDE or SCSI in the last year. Sure 
if you have the money to spend, SCSI have traditionally been more reliable.

>And IDE is NOT as fast as SCSI.  It probably never will be.  Sure, cache can
>make bursts on IDE work fast, and well.  But you load up an IDE drive, and
>it just can't put out like SCSI can.  That's especially true when there are
>two devices on a chain.  (Granted you pointed that out), but let's say I
>want an average server with half a dozen disks, in a RAID 5 Array (I'm no
>fan of RAID 5 either but it's sort of a neccessary evil) plus a hot standby.

Either way with a 8 drive array. I can max out the PCI-X bus with either 
IDE or SCSI drives. Both arrays will be faster than the system bus can 
handle. So there will be no advantage. Of Course I should know because we 
build systems like this every day.

>With SCSI, I plug it in, and go.

Same with 3ware cards.

>In order to do this with IDE, I need a seperate controller for each drive.
>If I add even just 6 or 8 drives, I have no PCI slots left in the server.
>(Once NICS, and a Remote Administration card is added.)  Then how do I
>expand it?

I only need 1 card for 12 devices. Have you checked out 3ware lately.


>I know it officially exists, but do you REALLY trust hot-swap IDE?  I've
>never tried it on IDE but I hardly trust it with SCSI.

Yes it works fine. It even works with software raid under linux.

>IDE needs a seperate chain for a few reasons.  IDE can't read and write at
>the same time, even to different devices if they are on the same chain.
>SCSI can.

Yes so SCSI is faster if you are talking about multiple devices on the same 
chain. A single SCSI drive against one IDE drive is not necessarily any 
faster if at all, it depends on application. If you are doing something 
that requires command queuing a single SCSI drive will be better.

>If you can accept the higher failure rate, and you don't expect much growth,
>the IDE can be OK.

The failure rate isn't that much higher any more.

>But for my money, your best solution would be duplexed SCSI drives.  And in
>my experience, I'd prefer Software duplexing, not hardware.
>It's a slight hit on the processor, that's true, but if it breaks, it's 
>easier to fix,
>especially as things get older, and hardware (proprietary RAID cards) become
>unavailable.  It's also DAMN expensive.  But it's more convienient than RAID
>5, and much faster too.  Especially during disk reads.

The raid 5 card you are using sucks. Yes raid 5 is more overhead but as 
long as you have a good enough processor it should be that much slower if 
at all.

Lots of Hardware Raid cards aren't any where near as fast as current system 
CPUs. I can guarantee better performance with Software Raid against all IDE 
Raid cards and most SCSI Raids. In fact some Hardware Raid cards are still 
a hit on processors. I have even seen IDE raid cards that use more CPU 
resources running Hardware Raid as opposed to Software raid. (And no I am 
not taking about the pseudo hardware raid cards like promise)

-- 
Mark Lane
Hard Data Ltd.
mailto:[EMAIL PROTECTED]

Telephone: 01-780-456-9771
FAX: 01-780-456-9772

11060 - 166 Avenue
Edmonton, AB, Canada
T5X 1Y3

http://www.harddata.com/
--> Ask me about our Affordable Alpha Systems! <--




BEGIN:VCARD
VERSION:2.1
N:Lane;Mark
FN:Mark Lane
ORG:Hard Data Ltd.
TITLE:Sales
TEL;WORK;BUSINESS:780-456-9771
TEL;WORK;VOICE:780-456-9771
TEL;WORK;FAX:780-456-9772
ADR;WORK:;;11060 - 166 Avenue;Edmonton;AB;T5X1Y3;Canada
LABEL;WORK;ENCODING=QUOTED-PRINTABLE:11060-166 Avenue=0D=0AEdmonton, AB T5X1Y3=0D=0ACanada
URL;WORK:http://www.harddata.com
EMAIL;PREF;INTERNET:[EMAIL PROTECTED]
REV:20010222T231737Z
END:VCARD

Reply via email to