Date: Wed, 16 Nov 2005 16:32:07 +0200 From: Danny Braniss <danny@cs.huji.ac.il> To: Brian Candler <B.Candler@pobox.com> Cc: freebsd-current@freebsd.org Subject: Re: iSCSI initiator driver beta version, testers wanted Message-ID: <E1EcOKV-000HR3-6B@cs1.cs.huji.ac.il> In-Reply-To: Your message of Wed, 16 Nov 2005 12:27:06 %2B0000 .
next in thread | raw e-mail | index | archive | help
Hi Brian, > On Wed, Nov 16, 2005 at 12:45:07PM +0200, Danny Braniss wrote: > > Haven't received much feedback, either it's a piece of s... or > > it's just working fantastic, or ? > > Anyways, I'm planning to be in Basel for BsdConn next week, so if anybody > > has something to say about it, i'm a good listener. > > I for one found the idea very interesting. If we could have an iSCSI target > too, then we would have a complete standards-compliant replacement for GEOM > ggated/ggatec, which might also form the basis of a low-cost SAN. [*] > > However I didn't reply, for two reasons: > > (1) I think you said your code had no error-recovery. In my experience, the > error-recovery code is typically the hardest to write, which implies to me > the code is a long way from complete and therefore not really worth testing. > That is, there's no point testing from the point of view "is this code ready > for production?" when code without error-handling isn't, by definition. > > That's just the impression I got of course, which may be wrong. > I guess I need to clarify some :-), by no error recovery, this is on the iSCSI side, the scsi part - via CAM - is fully error recovarable. the iSCSI protocol has error recovery too, and it's this one that's missing, ie: what happens if the TCP connection falls. so if your network is ok, and no one steps on cables, then it's working fine, no disk corruption, no data loss (FLW - Famous Last Words). i need to test the initiator with different/more targets, it seems that the RFC has more than one interpretation :-) > (2) I didn't fancy setting up a Linux box just so that I had an iSCSI target > :-) no need, we did, and we can crash it :-) > > Regards, > > Brian. > > [*] It might be necessary also to have some magic which maps arbitary > devices under the CAM layer, e.g. so you could share an IDE drive, or a > gvinum volume, as if it were a SCSI device. Or perhaps the iSCSI target can > just be a userland daemon, talking to the CAM layer where the device > supports it, or GEOM otherwise. correct. danny
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?E1EcOKV-000HR3-6B>