Date: Fri, 11 Aug 1995 12:32:23 -0700 (PDT) From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com> To: smp@teal.csn.org (Steve Passe) Cc: jmb@kryten.atinc.com, hardware@freebsd.org Subject: Re: Upgrade to my machine Message-ID: <199508111932.MAA04817@gndrsh.aac.dev.com> In-Reply-To: <199508111846.MAA28075@clem.systemsix.com> from "Steve Passe" at Aug 11, 95 12:46:39 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> > Hello, > > > talking of striping, a while back, you, terry and others (if i > >remember correctly) discussed disk striping and spindle syncing a > >number of drives together to produce a screamer disk system. was that > >code tainted in some way ?? No, the code is not tainted. The code is working for _RAW_ disks. I have done 2 to 4 wide raw strippes using spindle sync'd DSP3053L's and multiple NCR 810 controllers. Best numbers where four drives, four controllers at 11.2MB/sec to the RAW disk. > >or may it appear in 2.1 ?? No, defanitly _not_ 2.1, perhaps 2.2. I ran into a snap when I went to try and put disklabels on the striped raw device. It is called slice code :-(. I didn't want to deal with it at that time as the slice code was rapidly evolving. Now that the stripe code seems to have stabalized when I get some time I will demouthball my work and go back at getting a file system on one of these beasts. My first cut code was really just a hatched job on the BSD 4.4 Lite concat disk driver. After hacking that to kinda prove that it would work I started over from scratch and write a new implementation of concat with more options (striping is just a part of what you should be able to do with this type of code, mirroring is the other one :-). For any of you out there who have ever managed Auspex NFS engines, you'll know just about what my code should look like when it gets done from the users view point. The block level interleaving mechanism needs to be more tuneable than what either the concat code or Auspex offers. In the case of the concat driver it is hard coded in the kernel :-(, in the case of Auspex it is based on having a lot of smarts in the scsi controller so you don't need to tune this much. On FreeBSD since disk drives will have widly varied cache sizes and transfer characteristics this part of the system needs to be highly tuneable at volume creation time or you can't get good results. It took me a whole afternoon to find the ``right'' set of numbers to make the spindle synced 3053L's play well togeather, things like rotation offset also have to be set properly in the drive based on the blocking factors used in transfers so that blocks come under the heads immediatly after you issue the write :-) > > two sets of 3053's striped would be incredible....especially at > >under 40 cents a megabyte > > my "dream" system would use a 3940W to stripe pairs of spindle-synced > drives, am also interested if anything exists along these lines. Proof of concept model was up and running here on quad NCR810's 3 months ago. Raw disk only, but hey, it worked! > also, I seem to remember seeing an ad recently for a WIDE SCSI drive > that claimed to do internal head striping allowing it to approach > a sustained 20 MB transfer rate on the SCSI bus. anyone know what > drive this was, I can't recall... (old age is a terrible thing!) When you find this info let me know PLEASE!!! Now, take 4 of them and stripe it! > > Steve Passe > smp@teal.csn.org > -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Reliable computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199508111932.MAA04817>