Date: Sun, 29 Jul 2001 17:36:15 -0700 (PDT) From: Matthew Jacob <mjacob@feral.com> To: "Jose M. Alcaide" <jose@we.lc.ehu.es> Cc: <scsi@FreeBSD.ORG> Subject: Re: DLT 4000 throughput and cstream Message-ID: <20010729173112.C44279-100000@wonky.feral.com> In-Reply-To: <3B64A8AE.39629CBE@we.lc.ehu.es>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 30 Jul 2001, Jose M. Alcaide wrote: > Matthew Jacob wrote: > > > > [ -hardware trimmed- Only one list is needed ] > > Hum. I sent the message to -hardware _only_... Sorry- I probably missed which list it was. SCSI issues belong on the SCSI list. The header line says "DLT 4000" so the presumption here, and some mail stated, "problems with tape driver". > > > > > KB/t tps MB/s > > > > 60.59 22 1.31 > > > > 60.59 22 1.31 > > > > 61.05 22 1.31 > > > > 60.62 22 1.32 > > > > 60.59 22 1.31 > > > > ... > > > > This is about half the speed you should expect. > > > > Testing with dump, w/wo cstream, is not helpful. You need to test with a > > non-zero data generator: > > You are right. I also tested "cstream -i - ..." and "dd if=/dev/urandom ...", > with similar result. Anyway, I am interested in the dump case, since > it is the actual application. BTW, I also tested dumping the same > filesystem to a file on another disk, achieving 7-8 MB/s while > dump was in pass IV. It does indeed look like this *might* be a dump issue- but it also could simply be that any single threaded action for dump(8) is going to make for lousy performance. > > > yorp.feral.com > tape_pattern_tester > > tape_pattern_tester: [ -v ] [ -b blksize ] [ -r blocks per file ] [ -n > > number-of-files ] -f no-rewinding-tape-drive > > yorp.feral.com > tape_pattern_tester -v -b 32k -r 1000 -n 5 -f /dev/nsa0 > > .......Rewind Tape > > ........Write Pass > > EOT at File 4 Record 1000 Offset 32768 (163840000 total bytes written) > > Elapsed Seconds: 60; Data Rate: 2.6MB/s > > .......Rewind Tape > > .........Read Pass > > EOT at File 5 Record 0 Offset 0 (163840000 total bytes read) > > Elapsed Seconds: 72: Data Rate: 2.16667MB/s > > Hum... 32KB block size... is that size appropriate for DLTs? Some > people recommend big block sizes, but I think they are confusing > the device block size with the size of an intermediate buffer > between the input and output streams. DLTs seem to accept very > big block sizes; however, I cannot see any improvements for block > sizes bigger than 32-64KB, provided there is a big intermediate > buffer, of course. 32K seems about right. Each device is different though. > > Hey, that "tape_pattern_tester" utility seems interesting. Where could I > find it? Did you write it? I wrote it, yes. I'll probably put it in the source tree sooner or later. In the interim you can pick it up via BitKeeper from my toolkit repository @ bk://blade.feral.com:9003, e.g.: bk clone bk://blade.feral.com:9003 (see http://www.bitkeeper.com to get a copy of BitKeeper). -matt To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-scsi" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20010729173112.C44279-100000>