At 11:04 AM 10/24/2004 -0700, you wrote:
>>I was
>>under the impression that even a single drive that was
>>SATA would offer a faster data transfer than an IDE
>>drive w/an 8mb cache.
>
>SATA has big configuration advantages to PATA drives, but currently, only a
>tiny performance difference. It does, however, have potential for greater
>performance, which is why the industry is slowly making the move to PATA.
>But for now, you will not experience performance gains going from a
>comparable PATA to a SATA.

Mark is correct. One could write a pretty large article on all of this. The
problem is that you can't think of this as a bunch of black box components.
Just because SATA has a theoretical transfer rate of 150Mbytes/sec, it
means very little. There is a physical limitation to how fast the data
passes under the read heads in the disk. On modern IDE drives this is maybe
50 to 70Mbytes/sec max, this may be advertised on the drives specs as
"sustained" transfer rate. Thus a UDMA-100 interface is just as fast as
SATA. The only time you might achieve higher rates are small peak bursts
where the data happens to be in the drives cache. Current IDE drives tend
to have 8Mbytes of cache, but still, this assumes that all the data you
need happens to be in the cache, which statistically is fairly low. Even
so, at 70Mbytes/sec, 8Mbytes of cache is exhausted pretty quick. Even
maintaining sustained transfer rates requires that the data be arranged
contiguously on the disk, which again, for large files, may not be the case.

The other approach to get better performance is sometimes said to be using
RAID 0. By alternating data across multiple disks (similar to dual channel
memory), you aren't tied to the performance of a single channel. But even
if you could get sustained transfers approaching 150Mbytes/sec, the other
issue is the internal bus speed. Most disk controllers are simply PCI
devices, even if built on the motherboard. The typical 32bit 33Mhz PCI bus
has a maximum throughput of 133Mbytes/sec, which now becomes the bottleneck.

At least one architecture has promise to alleviate this somewhat, and that
is the Intel 865/875 chipsets using the ICH5 or ICH5R southbridge. The SATA
controllers on those are NOT PCI devices, and have availability to the
266Mbyte/sec bandwidth of the interbridge bus. The up and coming PCI
express standard can also overcome the PCI bus limitation.

I could go on about SCSI subsystems, 64 bit PCI controllers, etc. But that
could be for another discussion.


Russ Poffenberger
Credence Systems Corp.
[log in to unmask]

         PCBUILD maintains hundreds of useful files for download
                     visit our download web page at:
                  http://freepctech.com/downloads.shtml