NetNews Usenet Archive 1993 #3/NN_1993_3.iso/spool/comp/os/vms/22255

Internet Message Format | 1993-01-28 | 6.8 KB

Path: sparky!uunet!olivea!charnel!rat!usc!rpi!batcomputer!munnari.oz.au!manuel.anu.edu.au!sserve!actew.oz.au!ins_wayne
From: [e-mail] (Wayne Myles)
Newsgroups: comp.os.vms
Subject: Re: Striping and Shadowing
Message-ID: <[e-mail]>
Date: 28 Jan 93 23:01:33 GMT
References: <[e-mail]>
Organization: ACT Electricity & Water
Lines: 153

In article <[e-mail]>, [e-mail] (Geoff Hill) writes:

>     We are considering buying DEC's disk striping and/or shadowing
>     software and would be interested in other people's experiences
>     with these products.   Our VMS environment will probably be
>     a VAX 4000/100 with 4 (or possibly 5) SCSI disks.

  [Take all of this with a grain of salt; I'm a bit over-the-top
  when it comes to preserving my data! :-)]

  Call me paranoid, but I'm no fan of SCSI for coal-face
  operations; for small systems, yes, but not mission-critical,
  bet-the-company software.

  The basic problem is that few SCSI disks support all of DSA.  I
  seem to recall that DEC's later SCSI disks support forced-error
  flagging.  This, you want.  Also, SCSI disks tend to have
  different failure modes; the RA and RF disks I've seen fail,
  usually give warning.  All the SCSI disks I've seen fail are
  one-off catastophes.  No retries, no warnings, just crashes.

  Ask about the integrity characteristics of the disks you look
  at; how big are the ECCs?  RA disks are 170 bit, RF disks 264.
  An RA disk can correct 8 bad bytes in a block as a result.
  Compare this to a SCSI disk, and weep!

  Make your salescritters work for YOU for a change! :-)

  Keep in mind, also, a "feature" of Phase I shadowing, supposedly
  corrected in Phase II, that if a bad-block develops in the
  INDEX.SYS or BITMAP.SYS files, it will mark your entire
  shadowset as read-only!

  Phase I doesn't support SCSI anyway, from memory; that was a
  recent addition to Phase II.  Take a look at the release notes
  from VMS 5.4-3 onwards for copious detail; don't take my word
  for it.  There's some very subtle bug/feature behaviour too.

>    1) Disk Striping:
>         - how effective in improving disk I/O ?
>         - how reliable is it ?
>         - How easy is it to install and use ?
>         - any disadvantages ?

  I haven't played with it myself, but the reports I've heard are
  good.  The improvements seem to scale as expected, it's
  reliable, and easy to install.  Start with freshly initialised
  disks, if you can.  Regard this as hearsay.

>     2) Volume Shadowing:
>         - how effective in improving disk I/O ?

  It doesn't necessarily.  It is primarily an availability/
  reliability mechanism.  In fact, it slows down write operations,
  since ALL members must complete the write I/O before the shadow
  operation can complete.  Read operations will (theoretically)
  complete in 1/n th of the time, given n spindles.

  A good rule of thumb is that if your read-ratio (reads over
  writes) is less than 75%, you'll probably take a noticeable
  performance hit.  Above that, you'll probably notice an
  improvement.  We have some prime targets, with read ratios over
  90% and huge I/O rates.  Some day...   ($$$)

>         - how reliable is it ?

  Phase I? A pain (see above note about hard errors in critical
  files) but otherwise bulletproof.  Shadow-copies can really ruin
  your performance.

  Phase II?  See the VMS Release Notes.  There's some real
  doozies, like being able to mount the same disk into more than
  one shadowset!  Copy/Merge operations have less severe impact on
  the system/cluster.

  Another major plus: A Phase II shadowed system disk does not need to be
  shadow-copied every boot!  All members are mounted early enough
  that no merge is required under most circumstances.

  Generally, if you follow the rules, it is bulletproof.  Just be
  careful, and plan your virtual unit numbers ahead of time.

>         - How easy is it to install and use ?

  Very easy.  You won't believe how easy :-)

>         - can you have more than one shadow volume ?

  All our disks are shadowed in the main cluster.  Some of our
  other clusters have shadowed system disks.  Another benefit of
  Phase II is that it supports many more shadow units.  More than
  you could reasonably use.  Many of our disks are one-member
  shadowsets, which means that (given the right version of
  VAXsim-PLUS) you can configure "hot spare" disks.

  This is a BIG plus.  What it means is that you can mount a
  "spare" disk into a low-priority shadowset, and if/when another
  disk seems to be failing, VAXsim will automagically dismount it
  from one set, and mount it into the failing one.

  Then, your friendly local DEC Engineer can dismount the bad
  disk, fix it, remount it, and off you go - without taking the
  system/application down.

  We haven't quite got that working yet; It requires VAXsim V2.???
  for Phase II - it's on our list of things to do :-)

>     3) Mixture of 1 & 2:
>
>        If both Disk Striping and Volume Shadowing are installed,
>        e.g. a shadow set of 2 disks is striped across 2 other disks,
>         - how easy is this to set up ?
>         - would we get improved I/O from shadowed reads *plus*
>           improved I/O from striped reads ?

  Er, I hope you mean to make a stripeset out of shadowsets.
  I don't think the reverse is possible. I could be wrong.
  If I remember correctly, a stripeset is simply a collection of
  cylinders on a virtual/physical disk or disks that appears as a
  logical disk drive, with the sectors interleaved across the
  members.

  It should be no harder than remembering to mount the shadowset
  before the stripeset.  In theory, reads would be a bigger "win"
  and the writes at least as good.

  The main thrust of shadowing is reliability; of striping, throughput.

  You'll want to have the stripe members on different controllers,
  if possible.  Same with shadow members.  Stripe-members, because
  you're trying to increase throughput - using the same cable
  won't give you much.  Shadow-members, because you want
  reliability and availability; if one cable or controller dies,
  you want to keep rolling.

>     Will either of these products work with non-DEC drives ?

  Hmm. I don't see why not, but be careful of that "forced-error
  flag" support; if you don't have it, you could end up using
  corrupt data and not know it.  Try before you buy!

  Of course, it goes without saying that the members of a given
  shadow-set should be of the same type, and preferably the same
  rev disk & controller.  Belt, Braces and a piece of string, to be
  sure! :-)

  Good luck!

--
Wayne Myles               | Email: [e-mail], [e-mail]
System Manager/PostMaster |        [e-mail]  (all via UUCP)
ACT Electricity & Water   |        [e-mail]  (Internet)
GPO Box 366, Canberra ACT 2601, Australia.  Ph: [phone] (w) Fax: x3439