root@maadfa.ma.adfa.oz.au (god) (09/20/90)
We are about to stripe two 700 MB disk (with separate disk controller) together. However, we are not sure that the following procedures as defined in /usr/man/cat8/invol.8 is correct, and no one else in local Apollo Support Group can help us with this as they have never done this before. We would appreciate any comments. % invol option: 1 -m Select disk: [w=Winch|s=Storage mod|f=Floppy|q=Quit][ctrl#:][unit#] w0:0 How many disks will you be grouping together? 2 Enter the striping option. 1 (sector stripe) physical volume name: boot_disk Enter remaining members of disk group: Select disk: [w=Winch|s=Storage mod|f=Floppy|q=Quit][ctrl#:][unit#] w1:0 .....invol... } From /usr/man/cat8/invol.8 as shown in the paragraph below, we deduce } the following procedures have to be performed to finally stripe the } two disks as one. } > After you've created a multi-disk set with 1 -m, you must run > invol's option 2 to add logical volumes to the multi-disk set. > Invol will remind you that you must run it with this option. > Specify the primary disk of the multi-disk set (the first disk you > specified) to invol. If you desire a bootable volume, run invol > with option 8, as described above. option: 2 Select disk: [w=Winch|s=Storage mod|f=Floppy|q=Quit][ctrl#:][unit#] w0:0 .....invol... option: 8 Select disk: [w=Winch|s=Storage mod|f=Floppy|q=Quit][ctrl#:][unit#] w0:0 .....Striping Complete..... Many thanks in advance. ------- Patrick Tang Guan Yaw, Phone : +61 6 268 8882 Dept. of Mathematics, EMAIL-ARPA/CSNET : gyp@maadfa.ma.adfa.oz.au ADFA, Canberra, 2600. UUCP : ..!uunet!munnari!maadfa.ma.adfa.oz!gyp AUSTRALIA ACSnet : gyp@ccadfa.cc.adfa.oz
thoopes@diana.cair.du.edu (Tom Hoopes) (09/20/90)
When striping two disks on a DN10000 with two controllers, be sure to get the new fixed salvol from Apollo. Our disk croaked three times (before Apollo replaced one disk) and salvol couldn't save what was left over because of a bug. I believe the fixed salvol will be in the October bug tape.
krowitz@RICHTER.MIT.EDU (David Krowitz) (09/20/90)
My understanding of "invol" (and I will admit up front, that I have not made a multidisk volume) is that option 1 initializes the disk (or disks) as a single logical volume, and that you only need to use option 2 if you wish to split the disk (or disks of a multiple disk volume) into more than one logical volume. Option 8 is always needed to create a OS paging file on a disk that you wish to boot the system from. The "n" modifier to options 1, 2, and 3 (non-bootable volume) only means that there will not be a "sys" and "sys/node_data" and a boot block(s) put onto the logical volume. -- David Krowitz krowitz@richter.mit.edu (18.83.0.109) krowitz%richter.mit.edu@eddie.mit.edu krowitz%richter.mit.edu@mitvma.bitnet (in order of decreasing preference)
jose@Inference.Com (Jose Fernandez) (09/20/90)
When our DN10000 was delivered, the disks were striped and the OS was installed. Ignorant (or suspicious) of this pre-condition, I reinstalled from scratch, following the recipe from APOLLO's Installation doc. I don't recall the specifics (it's been over 6 months of a turrent over the dam) but I followed what seemed to be the obvious thing to do. My machine diskserves for all of our APOLLO development and so far, happy trails. This may message may not be very helpful -- it was meant to be encouraging. Good luck. Jose A. Fernandez Inference Corporation
thompson@PAN.SSEC.HONEYWELL.COM (John Thompson) (09/21/90)
> We are about to stripe two 700 MB disk (with separate > disk controller) together. However, we are not sure that the following > procedures as defined in /usr/man/cat8/invol.8 is correct, and no one > else in local Apollo Support Group can help us with this as they have > never done this before. We would appreciate any comments. > ... > The excised text seems correct -- it appears that you did the right thing, or that the example (if that's what it is) is correct. Here's the scoop, from someone who has striped disks many ways, many times -- Executing INVOL option '1 -m' does no formatting. What it _DOES_ do is set up some internal junk (with the Phys Volume label?) so that the node refers to ALL drives when you mention the primary. After you tell it what drive to format (e.g. w0:0) it'll ask how you want them striped, and how many drives you're striping together. After you feed it all the info, it does virtually nothing (for maybe 15 seconds?) and then tells you to execute option 2. Option 2 is normally used to add logical volumes to a single disk. However, now that there's disk striping, it really adds logical volumes to a single VIRTUAL disk (which is one or more PHYSICAL drives). The newly striped disk (I'll refer to each physical disk as a 'drive') has a huge vacancy -- the entire media is empty! Now you specify the volumes that will be on this disk (if you don't stripe, INVOL option 1 asks you this stuff), and how rigorously you want to format (no verification, format/write, format/write/read). Notice that, once you stripe drives into a disk, the multi-disk set is referred to by its primary number -- the first drive that you mentioned when you ran INVOL option 1. This referencing goes for INVOL, SALVOL, MTVOL, the Unix mount commands, .... There are 2 methods of striping disks -- 'cylinder' and 'sector'. If you have 1 controller per drive, sector striping will speed up disk access. If you have multiple drives per controller, sector striping will slow down access. The reason is fairly simple -- in sector striping, each drives cylinders are logically combined, with one drive having the even sectors and one having the odd (I'm assuming 2 drives). If you have 2 controllers, each one can ask for a sector, and you get back 2 sectors in about the same time you get one. If you have 1 controller and 2 drives, you get no possibility of parallel accesses (the controller waits for the drive to respond), and you get no possibility of multiple sectors being returned from a single disk access. You need to ask for drive 0's info, drive 1's, drive 0's, drive 1's, etc. With cylinder striping, you get no speedup/degredation regardless. In this case, one drive gets the even CYLINDERS, and the other gets the odd ones. Even with two controllers, you'll rarely need information that happens to be at the tail of one drive's cylinder and the head of the other drive's next cylinder. Be warned that, when you stripe drives, you're putting data on a disk that's twice as likely to fail. If either drive bites it, the entire multi-disk needs to be formatted. On the other hand, you can get higher throughput, larger capacity, more impressive 'lvolfs / df' responses, .... :-) Good Luck -- hope this helps. John Thompson (jt) Honeywell, SSEC Plymouth, MN 55441 thompson@pan.ssec.honeywell.com As ever, my opinions do not necessarily agree with Honeywell's or reality's. (Honeywell's do not necessarily agree with mine or reality's, either)
system@alchemy.chem.utoronto.ca (System Admin (Mike Peterson)) (09/22/90)
In article <9009201311.AA25490@richter.mit.edu> krowitz@RICHTER.MIT.EDU (David Krowitz) writes: >My understanding of "invol" (and I will admit up front, that I have not made a >multidisk volume) is that option 1 initializes the disk (or disks) as a single >logical volume, and that you only need to use option 2 if you wish to split the >disk (or disks of a multiple disk volume) into more than one logical volume. >Option 8 is always needed to create a OS paging file on a disk that you wish >to boot the system from. The "n" modifier to options 1, 2, and 3 (non-bootable >volume) only means that there will not be a "sys" and "sys/node_data" and a boot >block(s) put onto the logical volume. It is absolutely essential that you do an option '2' after building a striped set with option '1', otherwise you will not have formatted either disk. On our previous system, the disk block headers had to contain the proper information, which was only written at format time - does anyone know if the Apollo controllers require this information be generated across a striped set (i.e. can you format separately then stripe and expect it to work - my guess would be no, but ...)? On our system we have 4 760 MB disks on 2 controllers, and they are striped across controllers (w0:0 with w1:0, and w0:1 with w1:1). Here are the invol commands to do the striping and formatting. I strongly recommend that full write/read verification be done to try to find all the bad spots that can be found with invol. Note that we make each striped disk just one (very large) partition since we want maximum flexibility of user and temporary files (some programs create 500 MB temporary files). For the DN10000 system disk pair, the sequence is: ex invol 1 -m build a striped group w0:0 select controller 0, disk 0 2 select the number of disks to be striped 1 type of striping to be done (sector) u0 physical volume name (doesn't really matter) w1:0 select controller 1, disk 0 as the second disk y there is more to do 2 -b initialize logical volume w0:0 select controller 0, disk 0 3 select full write and read verification 8 use 8 K average file size all make each physical disk a logical disk volume y use the prerecorded badspot data y there is more to do 8 w0:0 select controller 0, disk 0 1 select logical volume 1 4096 enter the page file size (4096) n there is no more to do ex calendar w select Winchester disk ... enter the proper date and time answering the prompts For the DN10000 user disk pair, the sequence is: ex invol 1 -m build a striped group w0:1 select controller 0, disk 1 2 select the number of disks to be striped 1 type of striping to be done (sector) u1 physical volume name (doesn't really matter) w1:1 select controller 1, disk 1 as the second disk y there is more to do 2 -b initialize logical volume w0:1 select controller 0, disk 1 3 select full write and read verification 8 use 8 K average file size all make each physical disk a logical disk volume y use the prerecorded badspot data n there is no more to do -- Mike Peterson, System Administrator, U/Toronto Department of Chemistry E-mail: system@alchemy.chem.utoronto.ca Tel: (416) 978-7094 Fax: (416) 978-8775