Date: Sun, 20 Aug 1995 15:23:48 +0800 (CST) From: Brian Tao <taob@gate.sinica.edu.tw> To: Dmitry Khrustalev <dima@bog.msu.su> Cc: hackers@freefall.FreeBSD.org, Terry Lambert <terry@cs.weber.edu> Subject: Re: Making a FreeBSD NFS server Message-ID: <Pine.BSI.3.91.950820151646.17751x-100000@aries> In-Reply-To: <Pine.SOL.3.91.950820080008.552A-100000@sunny.bog.msu.su>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 20 Aug 1995, Dmitry Khrustalev wrote:
>
> Wrong, Sun had never included a method for enabling async writes in their os.
> ( other than adb, anyway). Ditto SVR4. SGI ships their boxes with async
> writes turned on by default, their reasoning is that every server is
> protected by UPS.
Blah. :( An interesting side note, comparing async NFS servers on
a 486 and an R8000-based SGI server:
(FreeBSD)
MB reclen bytes/sec written bytes/sec read
1 512 780335 729444
1 1024 838860 745654
1 2048 871543 758292
1 4096 766958 771366
1 8192 849479 749819
2 512 852176 628654
2 1024 794187 754032
2 2048 865920 693631
2 4096 780335 762600
2 8192 808540 754032
4 512 798915 586103
4 1024 772476 572967
4 2048 808540 725501
4 4096 822160 568719
4 8192 797727 603904
8 512 639512 659951
8 1024 836899 708272
8 2048 843473 584508
8 4096 851500 681308
8 8192 838860 643344
16 512 862789 642574
16 1024 833650 672981
16 2048 836573 549509
16 4096 839188 693407
16 8192 822790 701792
(PowerChallenge M)
MB reclen bytes/sec written bytes/sec read
1 512 729444 276737
1 1024 512281 322638
1 2048 571139 358870
1 4096 459649 358870
1 8192 695428 358870
2 512 686535 334290
2 1024 573580 256630
2 2048 618514 340654
2 4096 656321 361772
2 8192 684784 359833
4 512 631612 346815
4 1024 601199 357199
4 2048 608007 321095
4 4096 608697 345032
4 8192 562168 293532
8 512 580714 244811
8 1024 588029 351815
8 2048 641039 330382
8 4096 558077 335020
8 8192 559240 310061
16 512 590942 340870
16 1024 539840 320328
16 2048 563348 383205
16 4096 461526 392377
16 8192 534731 389530
In both cases, the CPU was running the same 100x100x20 cellular
automaton code, with ~2% idle time. NFS on the SGI slows down to a
crawl, especially on directory lookups, but the FreeBSD box putters
along at roughly the same speed, but at the expense of the simulation
execution speed. Does FreeBSD just happen to give disk-bound
processes higher scheduling priority than IRIX 6.01, and is this a
tunable paramter?
--
Brian ("Though this be madness, yet there is method in't") Tao
taob@gate.sinica.edu.tw <-- work ........ play --> taob@io.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSI.3.91.950820151646.17751x-100000>