Stephen> This "benchmark" is so full of variables that it is
Stephen> utterly meaningless on its own. You are running on two
Stephen> different systems. Is performance repeatable or not?
Stephen> How much memory is in each system? How fragmented are
Stephen> the existing disks? How much of the data is in the
Stephen> buffer cache when you do the rm?
I agree that it's utterly meaningless on its own, but one of the
complaints I receive most often about Linux is that deleting large
files/large directories takes a *long* time on many systems, which
suggests that there is a small problem hidden somewhere.
If there is a better benchmark that I can test with, please let me
know what it is, so that I can get more meaningful statistics.
I received several responses to my report in private mail (I wish
they'd Cc:ed linux-kernel) -- they confirmed what I'd noticed, that
SCSI behavior in Linux lately has gotten slower and slower.
From: "Daniel Roesen" <droesen@gmx.net>
Subject: AW: SMP 2.1.131: SCSI performance extremely poor vs. IDE
To: "Ben Gertzfield" <che@debian.org>
Date: 09 Dec 1998 08:58:33 -0800
I noticed this with an ICP Vortex RAID Controller (100GB Stripeset) attached
to a 4xXeon SMP, too. 5MB/sec with large blocks with bonnie. Not quite
impressive...
*snip*
Ben> This is, frankly, terrible. Since when is IDE three times
Ben> faster than ultra-wide SCSI?
Stephen> IDE command setup latency is a _lot_ lower than SCSI's,
Stephen> especially with UDMA. If all you are measuring is random
Stephen> read seek rates, then yes, IDE can quite easily be faster
Stephen> than SCSI.
But three times faster? Something's not quite right there. :)
Ben
-- Brought to you by the letters Q and S and the number 9. "Mmm.. slanty.." -- Homer Simpson Debian GNU/Linux -- where do you want to go tomorrow? http://www.debian.org/ I'm on FurryMUCK as Che, and EFNet and YiffNet IRC as Che_Fox.- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/