Date: Sun, 23 May 1999 12:45:19 +0200 From: Poul-Henning Kamp <phk@freebsd.org> Subject: UFS parameter survey: HELP WANTED! Message-ID: <8293.927456319.1@critter.freebsd.dk>
next in thread | raw e-mail | index | archive | help
------- =_aaaaaaaaaa Content-Type: message/rfc822 Content-Description: Original Message Subject: UFS parameter survey: HELP WANTED! From: Poul-Henning Kamp <phk@freebsd.org> Date: Sun, 23 May 1999 12:45:19 +0200 Message-ID: <8293.927456319@critter.freebsd.dk> Sender: phk@critter.freebsd.dk Bcc: Blind Distribution List: ; MIME-Version: 1.0 I recently made a 15GB filesystem and ended up with almost 500 cylinder groups. That is unlikely to be optimal. I talked to Kirk about the right parameters for UFS on modern disks some years back, and he said that no more than maybe a hundred cylinder groups made any sense. I think the fact that disks have gotten 25 times larger since the newfs paramters were last tweaked means that it is time to do so again. Unfortunately determining the parameters are not simple, so I would like to solicit help from as many as possible in determining if we should retune the newfs defaults. What I'm looking for is hard and soft data on the difference it makes for various sets of parameters, for various workloads and programs. So if you have time and facilities, lend me a hand. Basically, we can only sensibly compare data from the same hardware with the same workload, otherwise there are too many things to compare. Not all things can be measured precisely, but try to provide as much data as you can, and as good data as you can, ie: don't change controllers move partitions change BIOS settings without noting that you did so. The newfs parameters I would like to map out are: -a -b -c -e -f -i -m -t -u I'm generally interested in all impacts of this, but in particular if you can measure one or more of these specific parameters: read performance write performance create performance fsck time space wastage "other" I have no particular wishes for what program/application is used to excercise the system, but I would always prefer real-world over synthetic benchmarks. If somebody could measure news-server and web-server performance for instance it would be great. If anybody feels like making a structured benchmark script which just takes a device name as arg and runs some standardized tests that would be great too! Please report all results to <fs-data@phk.freebsd.dk> using this form. Put the information instead of the "___", but leave the line number intact please. You don't need to return the lines starting with # I will post news and updates about this project on: http://phk.freebsd.dk/ufs If there is sufficient interest we will make a mailing list too. Thank you for your participation! Poul-Henning *BEGIN UFSTUNE FORM* # Your email address. This will be used only to catalogue and # request further details from you. It will not be published # or distributed. # Example: # 1 phk@freebsd.org 1 ___ # Identity of the system you used. This is just to keep all measurements # straight. It is used with your email as a unique index. This # should identify one particular combination of hardware, excluding # the disk you had the filesystems on. If you have the disk on # different controllers in the same system, that will count as two # systems. Use names/numbers/whatever helps you keep track of things. # Please use the same thing for all measurements made on the same # system. # Example: # 2 rover using NCR controller 2 ___ # Identity of the disk/device you had the filesystem on, please # cut&paste the <...> piece from /var/run/dmesg.boot: # Example: # 3 <Quantum XP34300W 81HB> 3 ___ # Describe the nature of the test in one-line form. # Example: # 4 Time to fsck filesystem with all four 3.2 CD's loaded 4 ___ # Describe the nature and conditions of the test in # sufficient detail that somebody else can repeat it. # Example: # 5 Filesystem is newfs'ed and mounted. The four CDs from the FreeBSD # 5 3.2 release were copied in using "find . -print | cpio -dump XXX" # 5 where XXX is mountpoint/cd[1234]. Filesystem unmounted and run # 5 /usr/bin/time -l fsck /dev/rsd0c 5 ___ # You must repeat the rest of the form for each experiment. # document the newfs commandline used. Include a -s option here. # Example: # 6 newfs -f 2048 -s 30720000 6 ___ # document any mount options, kernel features or softupdates. # Example: # 7 softupdates 7 ___ # note any other detailes pertaining to this experiment (multiline) # Example: # 8 BIOS set to 5 MHz/narrow 8 ___ # document the result of the experiment, for instance the output from # time(1) or similar (multiline) # Example: # 9 2.91 real 0.03 user 0.05 sys 9 ___ *END UFSTUNE FORM* -- Poul-Henning Kamp FreeBSD coreteam member phk@FreeBSD.ORG "Real hackers run -current on their laptop." FreeBSD -- It will take a long time before progress goes too far! ------- =_aaaaaaaaaa-- To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?8293.927456319.1>