*BSD News Article 7596


Return to BSD News archive

Newsgroups: comp.unix.bsd
Path: sserve!manuel.anu.edu.au!munnari.oz.au!spool.mu.edu!sol.ctr.columbia.edu!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!ames!sun-barr!sh.wide!kogwy!math-keio!mad
From: mad@math.keio.ac.jp (MAEDA Atusi)
Subject: Re: bonnie i/o test results
In-Reply-To: eoahmad@ntuix.ntu.ac.sg's message of Sun, 8 Nov 1992 15: 26:11 GMT
Message-ID: <MAD.92Nov9195616@amber.math.keio.ac.jp>
Lines: 92
Sender: news@math.keio.ac.jp
Nntp-Posting-Host: amber
Reply-To: mad@math.keio.ac.jp
Organization: Faculty of Sci. and Tech., Keio Univ., Yokohama, Japan.
References: <CGD.92Nov6232822@eden.CS.Berkeley.EDU>
	<1992Nov8.152611.26176@ntuix.ntu.ac.sg>
Date: Mon, 9 Nov 1992 10:56:19 GMT

Sorry for those who are bored with disk benchmark stuff.  This is
(hopefully) my last article about on this topic.

In article <1992Nov8.152611.26176@ntuix.ntu.ac.sg> eoahmad@ntuix.ntu.ac.sg (Othman Ahmad) writes:

 >Chris G. Demetriou (cgd@eden.CS.Berkeley.EDU) wrote:

 >: PLEASE, before posting any more benchmarks:
 >: 	(1) learn about disk architecture.

At least, no one now dare to say "take away buffer cache from Unix fs".
A progress (somehow).

In article <1992Nov8.152611.26176@ntuix.ntu.ac.sg> eoahmad@ntuix.ntu.ac.sg (Othman Ahmad) writes:

 >No benchmark is perfect but at least we have figures. I'm fed up of people 
 >who only say but do not produce any figure. Those who say the most does not
 >mean that they are right.
[..deleted..]
 >	  Some e-mail me in saying that he can get 5Mbyte/second. It is faster
 >than any workstation that I've tested. However I do not trust it so much
 >until he posts the complete test details.

Ok, ok, here's figures.  I would not claim ANY conclusions.  JUST figures.

System:		PC Direct 486dxe
CPU:		486DX/33MHz
Cache: 		256K 25ns
RAM: 		8MB 70ns
Controller: 	Noname IDE
DISK:		Alps HD-A312C (200M IDE, 12ms)
OS:		Linux 0.98pl3
		Minix fs

Result of `iozone auto' from console (single user, no active background tasks).


	IOZONE: Performance Test of Sequential File I/O  --  V1.15 (5/1/92)
		By Bill Norcott

	Operating System: POSIX 1003.1-1988

IOZONE: auto-test mode

	MB      reclen  bytes/sec written   bytes/sec read      
	1       512     1777247             4032984             
	1       1024    2139951             4993219             
	1       2048    2231012             5518821             
	1       4096    2231012             5825422             
	1       8192    2231012             5825422             
	2       512     1613193             3813003             
	2       1024    1855886             4877097             
	2       2048    1906501             5377312             
	2       4096    1923992             5518821             
	2       8192    1941807             5518821             
	4       512     1421797             3554494             
	4       1024    1547713             4712701             
	4       2048    1625699             5053378             
	4       4096    1638400             5178153             
	4       8192    1632024             5309245             
	8       512     554435              59692               
	8       1024    559613              60271               
	8       2048    715751              113038              
	8       4096    1026757             192885              
	8       8192    562616              227025              
	16      512     509171              59741               
	16      1024    536184              61749               
	16      2048    563750              105570              
	16      4096    562239              183578              
	16      8192    568912              265294              
Completed series of tests

Now you can see the reason why I said "1MB is meaningless if you want
to measure disk performance".  In previous article, I (naively)
assumed this also holds on 386bsd.  265KB/s reading rate looks too
slow, but that's another story (I woudn't try to discuss here).

And `free' command just after the test prints:

             total       used      cache       free     shared
memory:       7376        908       6388         80        304
swap0:        8268          0                  8268          0


It means 6388 1k-pages are allocated for buffer cache.

;;; Benchmarking without analysis
;;;   is as useless as analysis
;;;     without benchmarking.
;;;
;;;		MAEDA Atusi (In Japan we write our family names first.)
;;;		mad@math.keio.ac.jp