[lug] 1T + SATA 300 performance

[Sean Reifschneider]

karl horlen wrote:
 > 232GB / 1.5 Gb per sec = 155 seconds to complete xfer
 > bits

First of all, your math is totally screwed up.  How much is one dollar
divided by 10 yen?  GB is something different than Gb, that's why they're
spelled differently.  One GB is 8 Gb.  So, 232GB / 1.5Gb is around 1200
seconds.  So, if your transfer was strictly limited by SATA speed, it'd
take around 20 minutes.

However...

Buying a controller that has a given rate does not magically make the hard
drive perform better.  You never mention the hard drive performance -- it
is not infinite...

So, you have to use the slower of the disc or controller or bus, etc, to
get the performance you can expect.  PCI is good up to around 100MB/sec, or
around 1gbps.  The controller is good up to 1.5gbps.  The disc is around
100MB/sec *AND* 10ms average access time.

This last number is very important.

So even at 1.5gbps, your hard drive is going to be the bottleneck, followed
closely by your PCI bus unless you have a 64-bit PCI-X slot (it's longer
than PCI, it'd be obvious if you had it).

But the access time really impacts the performance.  There are 100 seeks
available per second if they each are 10ms, so in other words every seek
costs you around 1MB worth of transfer.

Now, that said, your 500MB drive is probably only good to around 50MB/sec.
Again, around 10ms access time.

Ever file that needs to be copied involves several seeks.  So, if you are
copying a billion 1K files you're going to get extremely different
performance than if you are copying a single file of the same total size.

Doing many small files will absolutely kill your performance.

So it's not just a simple division, even if using the right units.

Sean
-- 
Sean Reifschneider, Member of Technical Staff <jafo at tummy.com>
tummy.com, ltd. - Linux Consulting since 1995: Ask me about High Availability