🚀 go-pugleaf

In demon.tech.pc, Butch Smith wrote:

>In article <34886102.16034927@news.demon.co.uk>, bob@lintilla.demon.co.uk 
>says...
>
>>>>Nope, the SCSI does bring a notable improvement.
>>>
>>>Care to back that up with some independently verified figures?
>
>
>Here are some figures to ponder.
>
>My threadmark results and system configuration:
>
>Data Transfer Rate (MB/sec) was 9.37
>Average CPU Utilization was 26.18 percent (using Adaptecs method of CPU 
>usage for every Megabyte/second of data moved this is 2.79%. see 
>--> http://www.adaptec.com/threadmark/ntcpubnch.html) 
>Shuttle 603 mb with 1 meg L2 cache 66mhz bus
>Intel P233mmx
>128 megs 60ns edo ram 
>Diamond Fireport40UW
>Seagate ST34501W Cheetah 4.5gb, Umax scanner, Plextor cd-rom, and Panasonic 
>cd-r on the scsi bus
>Running Windows NT 4ws with sp3 and IE4
>Test was run on the second of three 1.42 gb partitions

Thanks for providing some hard figures, though your system is towards the top
end rather than typical. For comparative purposes its not very useful as there
are no 10,000 RPM EIDE drives (marketing reasons, rather than technical?). Any
performance benefit you see will be almost entirely due to spin speed rather
than a better interface.

Threadmark results don't mean much when running applications, try the same thing
doing an Access query for example.

However, your figures are interesting in themselves. On paper the ST34501 should
achieve 11 to 17 MB/sec, average say 14 MB/sec. So you are getting 70% of the
expected rate, which probably is quite good depending on transfer size. 

In practice you can only get the paper transfer rates if you use large block
sizes with sequential reads. In typical applications doing mixed read/write
random access, small transfers, the max transfer rate disappears quickly. The
limiting factors are seek time and latency.

Regards,
-- 
Bob Cousins, Software Engineer.
Home page at http://www.demon.co.uk/sirius-cybernetics/