Building a new system: SCSI or IDE?

[Jonathan Sachs]

The OS should be able to tell you. If you can name the drives then I
could lookup the noiselevels for you, for comparison.

Do you know where? (This is Windows 2000.) I tried Device Manager and
Disk Management, but neither one was helpful. I used to have a SCSI
diagnostic utility that would display complete details on the SCSI
chain, I can't lay hands on it now.

My mail address is jsachs177 at earthlink dot net.

 

[Jonathan Sachs]

That's where I would look....

Ah, found it. Disk Manager displays the information if you select a
physical disk and open the Properties box.

Drive 0 is a Seagate ST336706LW.
Drive 1 is a Quantum Atlas V-18-WLS.

My mail address is jsachs177 at earthlink dot net.

 

[Folkert Rienstra]

Ah, found it. Disk Manager displays the information if you select a
physical disk and open the Properties box.

Drive 0 is a Seagate ST336706LW.

Cheetah 36ES 40/48/52 MB/s, 7 ms, 1.1 sone / 2.0 sone

Drive 1 is a Quantum Atlas V-18-WLS.

16/23/30MB/s, 13ms, 1 sone / 2.9 sone

Raptor, WD360GD, 34/46/55 MB/s, 7.9ms, 0.9 sone / 1.5 sone

 

[Mark]

Speaking of which, how fast is the latest SCSI? I know in its time, it was
great, since MFM and early IDE were nothing compared to SCSI but IDE has
been faster than SCSI (that i know of ) since around the time of ata33/66. I
could be wrong but are there even any new versions of SCSI out now?

Thanks

 

[Eric Gisin]

Nonsense. Back in 97 UDMA was 33MB/s and LVD SCSI was 80MB/s. SCSI needs more
speed because you put more drives on each cable. sATA is 150MB/s and SAS
(serial SCSI) will start at 300MB/s.

| Speaking of which, how fast is the latest SCSI? I know in its time, it was
| great, since MFM and early IDE were nothing compared to SCSI but IDE has
| been faster than SCSI (that i know of ) since around the time of ata33/66. I
| could be wrong but are there even any new versions of SCSI out now?
|

 

[Folkert Rienstra]

Nonsense. Back in 97 UDMA was 33MB/s and LVD SCSI was 80MB/s.

I don't think so. Maybe, with the greater part of a year between them.
c't lists UDMA-2 in late '79 but no Ultra2. SCSI faq says '98 for Fast40.

SCSI needs more speed
Bandwidth.

because you put more drives on each cable.

sATA is 150MB/s and SAS (serial SCSI) will start at 300MB/s.

Actually, comparing interface rates like that is 'nonsense'.
SAS uses dual 150MB/s ports that aren't necessarily used simultaniously.

 

[Folkert Rienstra]

Currently SCSI goes up to U320, which is 320 MB/sec, vs 133 max for ATA
and 150 for SATA.

The previous version was U160, which is still faster
than any current ATA variant.

Nope, you can't compare them just like that. For a fair comparison you need
to compare SCSI using 4 drives p/ channel with IDE using 2 drives p/ channel.

On a bandwidth per drive basis the ranking is
(1) SATA150, (2) SCSI U320, (3) UATA 133, (4) SCSI U160.

As for "in its time", SCSI is still preferred for servers, although there
are some at the exteme low end of the market now shipping with IDE drives.

Which probably don't serve a traditional multitasking/multiuser server role.

 

[Jonathan Sachs]

Currently SCSI goes up to U320, which is 320 MB/sec, vs 133 max for ATA
and 150 for SATA. The previous version was U160, which is still faster
than any current ATA variant.

No disk can keep up with even ATA100, so for workstations (not cheap
workstations; not low-end workstations; all realistically configured
workstations) this is academic. Even if you have two or three drives,
the practical difference will be small if it can be measured at all.

For servers with many drives and massively concurrent operations, the
extra channel capacity is useful.

For more than ten years, every workstation I have built for my own use
has had SCSI drives. I would describe myself as a SCSI bigot without
hesitation. But over the last couple of weeks I have concluded that
for workstations, the performance advantages of SCSI are no longer
meaningful.

My mail address is jsachs177 at earthlink dot net.

 

[Folkert Rienstra]

No disk can keep up with even ATA100, so for workstations (not cheap
workstations; not low-end workstations;

all realistically configured workstations) this is academic.

That depends very much on what you call 'realistic' and 'academic'. If that
is to mean that you won't use 2 IDE drives per channel, then you are correct.

Even if you have two or three drives, the practical
difference will be small if it can be measured at all.

That depends on which two sit on the same channel and whether them
two will be used concurently and how they will be used concurrently.

For servers with many drives and massively concurrent operations, the
extra channel capacity is useful.

I see no difference other than that a server may be busy _all of the time_
and a workstation not. There is only so much concurrent IO that you can
generate on a 2 drive (=IDE) channel.

In SCSI and IDE alike you choose your channel bandwidth on the num-
ber of drives that you plan to be using concurrently at _any_ moment.

For more than ten years, every workstation I have built for my own use
has had SCSI drives. I would describe myself as a SCSI bigot without
hesitation. But over the last couple of weeks I have concluded

I think the word is decided, not concluded.
Concluded means that based on the usage pattern you should go for SCSI.
Not because of channel bandwidth but because of your IO pattern and
the mechanical properties of the drives.

that for workstations, the performance advantages of SCSI are no longer
meaningful.

If they aren't now, then they never were.

 

[chrisv]

You like the moron word don't you. It must remind you of you.

Dave

No, it reminds me of top-posters, "Dave".