Date: Wed, 11 Apr 2001 15:43:15 -0700 (MST) From: Eric Lee Green <eric@estinc.com> To: "Kenneth D. Merry" <ken@kdm.org> Cc: Ken Menzel <kenm@icarz.com>, freebsd-scsi@FreeBSD.ORG Subject: Re: 'ch' Errors using chio w-sony changer Message-ID: <Pine.LNX.4.21.0104111522500.25769-100000@england.inhouse> In-Reply-To: <20010411151501.C1830@panzer.kdm.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 11 Apr 2001, Kenneth D. Merry wrote: > On Tue, Apr 10, 2001 at 15:50:48 -0700, Eric Lee Green wrote: > > though I have > > my doubts about the desirability of kernel-mode changer drivers in the > > first place, given the existence of pass<n> and the slow speed of > > changers, which makes a user-mode driver plenty fast... but then, I'm > > biased because I wrote a user-mode changer driver for FreeBSD that uses > > pass<x> :-). Since ch0 is part of the FreeBSD kernel, it probably should > > work right (shrug), even if it's not really necessary now that the pass0 > > device works (and works well, I might add). > > One advantage of having a kernel driver is the ability to cache state > somewhat. (With a userland program, unless you have some sort of config > file, you'd need to ask the changer about its configuration, etc., every > time you open the device.) Given how slow tape robotics currently are, that's pretty trivial overhead. I've had people use 'mtx' on 100+ tape libraries and they tell me it works fine, they just need to pipe the results of 'mtx status' through 'more' :-). It takes an Ecrix Autopack exactly 3 minutes and 35 seconds between the time you issue a 'mtx load' and the time that the tape actually comes online. I measured it because I was astounded at how slow it seemed to be (agh!). A few hundred milliseconds delay fetching some configuration info seems trivial beside that. This is also why I am not particularly worried about the fact that the storage management subsystem in BRU Professional is currently written in Python, not the world's fastest computer language. As slow as it is, it still takes far less time to decide which tape to place in the drive than the robot takes to actually move the tape to the drive. > > BTW, if you want to try it, 'mtx' works fine on FreeBSD. > > http://mtx.sourceforge.net . Also works fine on Linux, Solaris 8, Solaris > > 2.[67], and IRIX. Uses same syntax on all of these guys, just uses a > > different device name for the passthru driver (e.g. /dev/pass1 on FreeBSD, > > /dev/sg1 on Linux, some godaweful concoction on Solaris, ....). With the > > latest version 1.2.11, it's just a case of "./configure ; make ; make > > install", thanks to the wonders of GNU Autoconf. > > Sounds pretty cool. Alas, it turns out that I jumped the gun on 1.2.11 w/FreeBSD. 1.2.10 works fine, but I made some timeout changes that appear to have broke 1.2.11 on FreeBSD (I'm starting to make my timeouts adjustable so that, e.g., a 'mtx inquiry' uses a short timeout and a 'mtx inventory' uses a VERY long timeout). The odd thing is that I tried 1.2.11pre3 on FreeBSD and it worked fine, so apparently I broke it on 1.2.11pre[456] which I did not try on FreeBSD (sigh!). Guess 1.2.12 will come out Real Soon Now with fixes for FreeBSD and IRIX. -- Eric Lee Green eric@estinc.com Software Engineer "The BRU Guys" Enhanced Software Technologies, Inc. http://www.estinc.com/ (602) 470-1115 voice (602) 470-1116 fax 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?Pine.LNX.4.21.0104111522500.25769-100000>