Conference decwet::networker

Hi,

  The current NT server does not have the big performance enhancements
that are being worked on.  A specific schedule is not agreed upon for
the release that has it (some time in '97, just when exactly).  Please
sign up for field test if you want to check out the next release
(http://www.service.digital.com/sms/nsrnt).

However from other data I've seen lately, one backup at one company got
about 900GB in 20 hours (4100, 4 CPUs, all local storage, ~1GB memory)
which roughly beats your need of 40GB/hour with the current product, but
I'm not sure how you'd break up your backup.... would node A only backup
local storage or do node B's data as well, and same question for node B
with respect to node A's data....  Anything going over the network will be
slower than that which is local, even after performance enhancements
are put in place.  Either way it is really important how the backup is
broken up.  Will you have additional time to use if you want to verify
your backup or clone your tapes?

  Please note the other topics for NT Clusters, topic  86.  That may
answer whether you want to use NetWorker or what questions you should
ask before committing yourself.  

    Since Digital Clusters for NT does not support tapes on a shared bus,
    you're looking at a solution which will not be available if the system
    with the libraries goes down.  Also, backed shared storage would always
    have to belong to 1 specific cluster member at backup time; otherwise,
    NetWorker would not be able to restore the data correctly since it
    might be backed up by different clients and the index information would
    be split between the clients.  Specific support for NT cluster disk
    services by NetWorker remains to be developed (something similar to the
    UNIX cluster client is needed).
    
    Microsoft clusters (Wolfpack) does not currently plan on supporting
    tapes on a shared bus either in its first release (this summer); so the
    above problems also apply to it.
    
    Mark
    Digital NT Clusters team

After thinking on this for a bit... if they had one TL812 (prefer TL894) on node A
and another on node B, i.e. have both systems be NetWorker servers, one could
clone tapes/saves after a backup on one server and put them in the other
library for complete availability.  Does require extra time for cloning not part
of initial window and does kinda assume that shared disks are equally distributed.

    
              <<< DECWET::DOCD$:[NOTES$LIBRARY]NETWORKER.NOTE;1 >>>
                                 -< NetWorker >-
================================================================================
Note 603.3                High perfomance backup for NT                   3 of 3
DECWET::LENOX "a bit too exciting"                    6 lines  21-APR-1997 16:24
                   -< for a really redundant situation.... >-
--------------------------------------------------------------------------------

    After thinking on this for a bit... if they had one TL812 (prefer
    TL894) on node A and another on node B, i.e. have both systems be
    NetWorker servers, one could clone tapes/saves after a backup on one
    server and put them in the other library for complete availability. 
    Does require extra time for cloning not part of initial window and does
    kinda assume that shared disks are equally distributed.

    Thanks for your comments,
    
    Node A will be doing CC-MAIL services (50 GB of disk).
    Node B will be doing File-Print services (160 GB of disk).
    
    In normal operation mode, disks mounted by server A will not be mounted
    by server B, and viceversa. If a failure happens the survival server
    will restart all services, mounting related disks. All applications
    issues will be properly exposed to the customer, in relation with
    restarting CC-mail.
    
    From your comments, I guess each node will be a Networker
    Server, each one will backup its locally mounted disks. Node A would have
    its own TL812. Node B would have two TL812s. If node B fails, backup
    window will be larger.
    
    Which is the behavior of Networker for File Services, can this backup
    be done on-line (with clients connected and using them)?.
    
    Same thing for CC-mail, can we backup CC-mail fails on-line, or we
    need to stop mail service?
    
    Can anybody explain a little bit what tape clonning is? Or give me a
    pointer to learn about it.
    
    Sorry for questions that could be basic, this is a very good
    opportunity where the large perfomance requirements could give us
    a differentiator against COMPAQ. The other approach is to go with
    more (than 2) tiny servers and COMPAQ would fit nicely.
    
    Thanks for youe help,
    
    Ivan. 
    
    

NetWorker will not backup files that are open and locked.
NetWorker will report if a file size changed during backup
and it is assumed (don't know where this is noted in the
docs) that the file will be backed up the next time, i.e.
don't assume a file is backed up ok if the file size
changes during backup.  If you want to back up open and
locked files, I suggest you try out St. Bernard's software
(www.stbernard.com) and see if that works out for your
situation.  They used to have trial software for NT Alpha
available on their web page.  Otherwise, file backup should
be ok.

I suggest that you don't put more than 2 drives per scsi
bus.  Depending on your filesystem layout and how you
do your group and client definitions, you may not get enough
streams to drive 8 tape drives with the current product.
That is worth looking into further.

Tape cloning is when one gets a copy of all the savesets
on a tape to one or more new tapes (since sometime the
tape you clone has savesets that span other tapes).
Saveset cloning is when on puts a copy of a saveset on a
different tape.  Cloning requires at least 2 drives.

plan ahead. Current tape drives cannot keep up with planned performance
 increases in NetWorker. For example:

  TZ87 = ~1.2-1.5 Mb/sec  (*3.515 = 4.2-5.3 Gb/hr)
  TZ88 = ~1.5-2.5 Mb/sec  (*3.515 = 5.3-8.8 Gb/hr)
  TZ89 = ~5 Mb/sec       (Digital Internal) (17.6 Gb/hr)

 The performance numbers we are talking about (see the White Paper published
 by the NetWorker Team) are in about a 300Gb database backing up to drives 
 recording 11Mb/sec (!!).  NOTE: these are uncompressed numbers as they are 
 more linear than compressed numbers. Be careful about that too!!

 most people plan only on getting NetWorker, and then wonder why they
 cannot back up their database *fast* - always look for the bottleneck in the
 data path. Find it, then fix it.

 If Oracle only delivers 1 backup stream per tablespace, this could limit
 how fast (ie, in parallel) you can get the job done. Alternatively, if
 you can get 4 streams going (for example) from the raw devices that
 compose that tablespace, then perhaps you can get the backup done faster
 (this example works in Unix Oracle, dunno about NT Oracle).

food for thought. Revenue for consulting services.

Open File Manager from St. Bernard software allows NetWorker to 
backup open files on NT.  OFM is available for NT Alpha.

We haven't tested this using NetWorker for NT Alpha, but Legato 
resells this product for their NT Intel platform.

Wes