Michal Jaegermann wrote:
>
> On Thu, Nov 30, 2000 at 11:49:33PM -0600, W Bauske wrote:
> > One of my 80GB 5400rpm
> > drives under 2.4.0test10 on an Athlon box:
> >
> > /dev/hdf:
> > Timing buffer-cache reads: 128 MB in 0.83 seconds =154.22 MB/sec
> ^^^^^^^
> This number is not really about a _disk_ performance at all.
>
I always run both tests. It is influenced as you describe
below with cache, which is the intent of the test.
> > Timing buffered disk reads: 64 MB in 2.25 seconds = 28.44 MB/sec
>
> And the other one should be treated carefully as well. 'hdparm -t'
> is a pretty primitive test and disks have caches too. I am not saying
> that it is not useful, far from it, but to consider it as an ultimate
> indicator of a disk performance in your _system_ is a mistake.
>
> > Also read a 10GB file in 367 seconds.
>
> And how often you are doing that?
All the time.
I have multiple TB's of UATA66 or better disks for holding
temporary datasets.
> Some might, but this is not a normal
> usage pattern.
Of course it depends what you use the machine for.
I use my file servers for lots of big files.
Filesystem 1k-blocks Used Available Use% Mounted on
/dev/hda3 1035692 883728 99352 90% /
/dev/hda1 69973 6877 59483 11% /boot
/dev/hda4 57842364 52646300 4020752 93% /z/wsb50a
/dev/hdb1 59068896 42040812 15827868 73% /z/wsb50b
/dev/hdc1 59084932 48969256 8915140 85% /z/wsb50c
/dev/hdd1 59084932 7403768 50480628 13% /z/wsb50d
/dev/hde1 59084932 5579916 52304480 10% /z/wsb50e
/dev/hdf1 59084932 10142580 47741816 18% /z/wsb50f
/dev/hdi1 59084932 33650420 24233976 59% /z/wsb50i
/dev/hdj1 59084932 26343792 31540604 46% /z/wsb50j
/dev/hdk1 59084932 28218508 29665888 49% /z/wsb50k
/dev/hdl1 59084932 30731672 27152724 54% /z/wsb50l
> How about a random pattern of reads and writes from
> /dev/hda to /dev/hdb and back? Yes, deliberately on the same channel
> and carefully enough to nullify effects of caches?
>
A 2MB cache on a drive will not effect significantly,
either the hdparm result or my real file tests. I/O
to the same channel has more effect than just cache.
Same thing occurs when you hit a SCSI bus with more than
one active device. IDE is no different.
> What I am trying to say is that, as old participants of this list should
> know, bare numbers are frequently misleading. :-)
>
I get this level of performance on UATA66/UATA100 drives
all the time. It's no fluke. I ran the tests after I read
peoples replies. My current limit is actually network
bandwidth. GigE is still to pricey to bother with so my
machine to machine transfers sit in the 9MB/sec range.
Wes
_______________________________________________
Axp-list mailing list
Axp-list@redhat.com
https://listman.redhat.com/mailman/listinfo/axp-list
This archive was generated by hypermail version 2a22 on Tue Jan 2 07:48:31 2001 PST
Send any problems or questions about this archive to webmaster@alphalinux.org.