🚀 go-pugleaf

In article <2134@acf5.NYU.EDU>, liuyu@acf5.NYU.EDU (liuyu) writes:
>ajb8886@ritvax.isc.rit.edu writes:
>
>>Also note that the "CMOS" drive configuration (what the computer THINKS the
>>drive is configured as) and what the drive is ACTUALLY configured as (due to
>>"translation"), the test may be actually spanning multiple cylinders depending
>>on the data on the drive is organized.  In such a case you would have more
>>overhead due to the head repositioning (which the diagnostics program "thinks"
>>is not occurring due to sector translation).
>
>I used the parameters on the label of this drive. It has 989 cyl., 12 head,
>and 35 sector/trk. Now come to think of it. It is unlikely to have 6
>platers (12 heads) in this 1" height drive. The *actual* geometry of this
>drive probably has 2 or 3 plater, more cylinders, and more sectors on outer
>tracks and less on inner. I guess total # of sectors is the same for both
>the parameter on the label and the *actual* parameter. Am I right?

Nope.  With most of the current crop of 1" high IDE drives, the actual number
of sectors per track varies from the innermost to the outermost cylinders. 
Typically between 30 and 38 sectors per track depending on the circumference of
the given cylinder.  The cylinders with highest sector density will give the
highest transfer rates and vice versa, because the drive is still spinning at a
constant rate (3600 RPM).

As for the number of platters, most 1" drives have two (four heads).  Typical
number of tracks per platter is often greater than 1200 cylinders.


>
> >Moral of the story:  Do not rely on disk benchmarks too heavily, as the way
> >they are measured can sometimes give widely varying results (particularly with
> >the use of on-drive hardware caches).
>
>I got ya.
>
> >Yes.  Try a recent version of CORE test.  It's pretty consistent ALTHOUGH it
> >sometimes gets confused by caching and reports very high performance (but
> >misleading) figures with some caches enabled.
>
>I just tried CORE test (v 2.92) I found on wuarchive. It showed 1.89 meg/s. 
>Amazingly different... Maybe this version is too old. It is dated 1990.

Core test shows the raw transfer rate as opposed to the FILE transfer rate. 
I've found it to be a better (if still not totally reliable) benchmark than
the hopelessly useless and innacurate disk benchmarks as supplied by QAPLUS 
and CHECKIT.

One word of warning : with SOME disk caches, the transfer rate with core will
report unreasonably high figures (like 10mb/sec :)  Interestingly enough, this
is the case with the old SMARTDRV.SYS but not the current SMARTDRV.EXE.  

Go figure


>
>Anyway, thanks for the detailed reply!


No problem.

Alex