On Nov 20, 2003, at 12:59 PM, Martin Heroux wrote:
However what I am surprised is that modern SCSI drive turning at 7500
RPM
are quite faster than 1Gbps card isnt ? So why does the storage unit
can be
a bottleneck ?
A Seagate Cheetah 15K.3 at 15,000 RPM:
http://www.seagate.com/docs/pdf/datasheet/disc/ds_cheetah15k.3.pdf
has an external transfer rate of 320MB/s (2.56Gb/s) when using Ultra
320 SCSI and 200MB/s (1.6Gb/s) when using 2GB Fibre Channel.
*However*, the internal transfer rates are 609 to 891 Mb/s, the
internal formatted transfer rates are 57 to 86 MB/s (456 to 688 Mb/s),
and the sustained transfer rates are 49 to 79 MB/s (392 to 632 Mb/s).
I don't know what the difference between "internal" and "internal
formatted" is, although the first of them might be "how fast you can
transfer bits from the head to the platter" and the second of them
might be "how fast can you transfer bytes of data from the head to the
platter", which might exclude
bytes that aren't data (sector headers, ECC, etc.);
encoding an 8-bit byte in more than 10 bits;
etc..
"Sustained" might include head switch times, track-to-adjacent-track
seek times, etc..
(I'm not a disk drive expert, so those are just guesses.)
So, just because the drive electronics, and the bus being used to
transfer data to the drive, can transfer some number of bits per
second, that doesn't mean you can keep moving those bits to the disk
surface at that rate (it might be transferring to or from a track
buffer, for example, but when the buffer fills up, you can't just keep
transferring).
And, over and above that, you might have:
file system overhead (allocating blocks to which to write, updating
the data structure that records the length, etc. of the file, and so
on);
bus bandwidth issues (1Gb/s of network traffic + 1Gb/s of disk traffic
is 2Gb/s, or 250MB/s - that's almost twice what a 33 MHz 32-bit PCI bus
can handle, although a 66 MHz 32-bit PCI could handle it as long as
there's not too much overhead or *other* traffic on the bus, and faster
or wider buses should have no problem);
bus *latency* issues (bandwidth is all very well and good, but if you
can't get onto the bus before your buffer fills up, the fact that
you'll have more than enough bandwidth later won't help you).
