Date: Tue, 3 Dec 1996 16:28:21 -0600 (CST) From: Joe Greco <jgreco@brasil.moneng.mei.com> To: mdtancsa@sentex.net (Mike D Tancsa) Subject: Re: News machine FreeBSD config Question Message-ID: <199612032229.QAA27483@brasil.moneng.mei.com> In-Reply-To: <199612032152.QAA25764@granite.sentex.net> from "Mike D Tancsa" at Dec 3, 96 04:52:08 pm
next in thread | previous in thread | raw e-mail | index | archive | help
This will be a terse response :-) Just because I don't have lots of time. > Hello, I am trying to configure our news machine for optimal performance running > FreeBSD 2.1.5R and I have a few config type questions I hope you can help me > with... > > > (*NOTE I have searched through the FreeBSD mailing lists and done a few > searches on DejaNews and did not find all the answers to my questions. If > however all this is in an FAQ somewhere that I missed, sorry to waste time > and could you please point me to the right document.*) > > The hardware is the following > > Pentium 133 with 128Meg of EDO RAM Good > NE2000 NIC Bad > AHA 2940 SCSI adapter Expensive but good > /dev/wd0a 31775 15651 13582 54% / > /dev/wd0s1e 1440639 1105935 219453 83% /usr > /dev/sd0s1e 998879 246095 672874 27% /usr/local/news > /dev/sd1e 3868777 2381285 1177990 67% /usr/local/newsspool > This is the main news spoool > > /dev/sd2e 580303 153198 380681 29% /usr/local/newsspool/misc > kinda a waste right now, but will change it... > > /dev/sd2f 628662 379097 199272 66% /usr/local/newsspool/rec > /dev/sd2g 785902 158492 564538 22% /usr/local/newsspool/over.view Don't put multiple partitions on a disk. You are _forcing_ the disk to seek all over the place. > I am starting InterNetNews server INN 1.4unoff4 05-Mar-96 ready with the > following... > unlimit datasize > unlimit stacksize > limit openfiles 1000 > limit maxproc 400 > > We are not really feeding anyone else save a few UUCP customers who have > perhaps 10-15meg a day that they slurp down from us. In terms of NNRP > connections, we have perhaps 10 or so at peak times... Our active file > contains around 16,000 groups of which in one week only 1500 were only > accessed in a one week period :-( Typical. > My questions are > > 1: I will be adding another Hawk SCSI 1 gig drive... How should I make use > of it? put the comp.* on it or use it for logs? Put overview on it, that's what I would do. Get rid of misc and rec partitions. My news spools look like this: % df -k [...] /dev/ccd1e 1971087 771521 1041880 43% /news /dev/ccd2e 1971087 1032440 780961 57% /news/.0 /dev/ccd3e 1971087 603507 1209894 33% /news/.1 /dev/ccd4e 1971087 610164 1203237 34% /news/.2 [...] % ls /news/.2 aus chinese de fj news relcom uk bit comp es junk rec tw us % ls -l /news/aus lrwxrwxr-x 1 news news 6 Dec 3 08:33 /news/aus -> .2/aus % This way multiple hierarchies may share the disk without having to force the drive to go seeking from here to hell. > the IDE is currently being > used just to spool /var/log/news and so on... watching vmstat, there is not > a lot of activity on IDE however. Will getting rid of the IDE make a > significant difference? Probably not, but.... ecccccccchhh. :-( > 2: Should I get rid of the NE2000 for a PCI based NIC ? Again, will it make > a big diff? If so, what is the PCI NIC of choice for FreeBSD ? Probably. NE2000 is about as bad as it gets. I know they're only $15 or so for clones. I ran a larger news server than yours for quite some time on a WD8003E, but that does not mean it is wise. The Kingston KNE-40T's wholesale for around $35-$40. Excellent performance (easy to saturate a wire) and low cost make this my PCI Ethernet controller of choice. > 3: Kernel configs.... Currently I have the following options... > ---------------------- > #options MATH_EMULATE #Support for x87 emulation > options INET #InterNETworking > options FFS #Berkeley Fast Filesystem > options NFS #Network Filesystem > #options MSDOSFS #MSDOS Filesystem > options "CD9660" #ISO 9660 Filesystem > options PROCFS #Process filesystem > options "COMPAT_43" #Compatible with BSD 4.3 > options "SCSI_DELAY=15" #Be pessimistic about Joe SCSI device > #options BOUNCE_BUFFERS #include support for DMA > bounce buffers > options UCONSOLE #Allow users to grab the console > > maxusers 64 > options "CHILD_MAX=128" > options "OPEN_MAX=1024" > options EXTRAVNODES=1 > > #the extravnodes I saw in another mailling list post and havent tried it yet... > > options SYSVSHM > options SYSVSEM > options SYSVMSG > options MAXMEM=131072 > ---------------------- > > Should I increase the maxusers even higher ? Are there any other kernel > settings (undocumented or what not) that I should fiddle with? I usually use maxusers=#MBofRAM. 64 should be OK though. > 4: Innd doesnt *seem* to be taking advantage of the memory I have in the > system... Its only about 18meg right now as reported by ps... > news 20373 9.0 13.7 18372 17688 ?? SNs Sun09AM 250:23.06 > /usr/local/news/etc/innd -p7 -i0 It will eat more as time goes on :-) I have seen innd actively occupying 160MB of RAM. load averages: 1.47, 1.24, 0.90 16:22:40 63 processes: 4 running, 59 sleeping Cpu states: 64.7% user, 12.4% nice, 17.8% system, 5.0% interrupt, 0.0% idle Mem: 213M Active, 1524K Inact, 18M Wired, 15M Cache, 7976K Buf, 300K Free Swap: 614M Total, 69M Used, 545M Free, 11% Inuse PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND 29853 news 105 0 44M 44M RUN 5:03 71.14% 71.14% expire 18076 news 51 4 109M 107M RUN 961:09 9.54% 9.54% innd 11817 news -14 4 7204K 7112K ufslk2 12:24 6.41% 6.41% innfeed 11818 news -6 4 5896K 5796K RUN 10:16 3.93% 3.93% innfeed 11819 news 2 4 3960K 4040K select 5:39 1.53% 1.53% innfeed 11820 news 2 4 33M 31M select 3:23 0.72% 0.72% innfeed This is newspump.sol.net. 256MB RAM, usually more cache but right now an expire is occupying some valuable RAM. > Top reports the following info > load averages: 0.08, 0.28, 0.23 > 15:47:27 > 43 processes: 2 running, 41 sleeping > Cpu states: 0.9% user, 0.9% nice, 7.8% system, 0.9% interrupt, 89.5% idle > Mem: 41M Active, 2068K Inact, 13M Wired, 70M Cache, 4855K Buf, 176K Free > Swap: 144M Total, 64K Used, 144M Free That's an excellent cache value. I am used to seeing that on a machine with twice the memory you have. Note: readers, feeds, and INND itself will slowly eat away at that. > PID USERNAME PRI NICE SIZE RES STATE TIME WCPU CPU COMMAND > 20373 news -6 4 18M 18M biowai 250:30 8.13% 8.13% innd > > 5: As you can see, innd has reniced itself to 4... Should I fiddle with this > value? If so, how do I prevent the spawned nnrpd's from also getting a > higher priorty? See config.data > 6: Has anyone using FreeBSD upgraded to INND 1.5? Is it worth it to upgrade > to it? Yes, and only if you need the features, in that order. > and... > > 7: Should I go through our active file and nuke groups that people really > are not reading? Will it make a big difference to go down from 16,000 to > about 10K? YES, this is a real win particularly once you get more readers. Active file size directly drives a lot of memory usage in INN. > 8: At what point should I add another SCSI controller ? Yesterday. Get several NCR-810's. They wholesale for about $65, I can buy three for what you paid for your 2940, and they work as well as the 2940. More importantly: you should get more disks. You can never have too many disks. My new "news.sol.net" will have 30 on six SCSI chains. > Thanks for looking this over and I appriciate any insights / suggestions you > can provide, Sorry for the quick reply, usually I try to explain "why" in more detail. ;-) ... Joe ------------------------------------------------------------------------------- Joe Greco - Systems Administrator jgreco@ns.sol.net Solaria Public Access UNIX - Milwaukee, WI 414/546-7968
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199612032229.QAA27483>