This argument gets older and more tiring every time I hear it.
I seriously wish that:
a) SCSI-proponents would stop trying to justify spending the money they've
spent on SCSI disk subsystems. You spent good money on good equipment. Be
happy with it.
b) IDE-proponents would stop trying to pretend that they are getting
anything other than a cheap, commoditized, consumer-oriented solution.
c) Both sides would start using the corret argument. It often seems that
SCSI advocates are proposing a high-performance server solution for light
workstation use, and IDE proponents are pushing IDE as having performance
comparable to SCSI in a server environment.
Both interfaces have their place in the computer world -- it's pointless to
argue that one is better than the other unless you know what the application
is.
On Fri, Jan 07, 2000 at 08:21:42PM -0500, Mark Hahn wrote:
> > once you get them under heavy load. All of the boxes with IDE get very
> > choppy and slow feeling. Regardless of the maximum burst rate an EIDE
> > controller or disk can support, they're just slower than SCSI. Any of the
>
> only true for misconfigured ide.
> "feel" depends mainly on whether you're busmastering or not.
SCSI drives are optimized for heavy usage. They will perform better while
being "thrashed" than their IDE counterparts. This is mostly due to the
tagged command queue, which can fetch blocks in a different order than they
were requested.
> > LVD SCSI disks can compile a full kernel in about 2:10, and is still
> > responsive under load. Putting IDE disks on any of those machines kills
> > their responsiveness though.
>
> ultra2 scsi is wonderful. it's also expensive. so much so that people
> who buy it sometimes feel obliged to stretch the truth about ide.
Let's try the opposite spin --
"ultra2 scsi is wonderful. it's also expensive. so much so that people who
don't want to spend the money feel obligated to stretch the truth about IDE"
> yes, you can easily achieve 2:10 on a modern ide system (udma).
Kernel compile times depend on a variety of factors, the most important of
which being memory size and processor speed. Unless you're using a processor
that can compile the kernel faster than it can be read off of disk, this will
not change.
> basically, modern ide disks sustain >20 MB/s.
> most modern scsi do better, up to around 40 MB/s.
These numbers are flat out wrong. In general SCSI and IDE drives (per RPM)
have about the same typical sustained transfer rates. (around 13 MB/sec
@7200 RPM.) The difference is that:
a) SCSI supports tagged command queueing
b) (Wide) SCSI supports 15 devices per channel.
c) SCSI uses less CPU overhead (yes, it does.)
d) SCSI is more widely supported
e) SCSI HBA's have their own BIOS, which enables SCSI drives to be free of
problems caused by buggy motherboard BIOS.
f) SCSI drives are available at 10k RPM.
Most of these advantages come into play when the system in question is going
to be a heavily accessed file server. For normal workstation use, the user
will have to make his own choice, depending on his power (and possibly
fiscal) level.
So, to sum up --
- People who use SCSI when IDE would suit them just fine are wasting money.
- People who aren't willing to pay for SCSI when they require high
performance are going to regret it (and probably will wind up buying SCSI
anyway, eventually)
> ide channels cost around $15 apiece.
> ide disks are at around $8/GB (verus ~$30/GB for scsi)
Those prices sound about right. However, anyone who puts price above all
other considerations is likely to regret many of their choices somewhere down
the line.
--Adam
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/dmentre/smp-howto/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
