On 29 Sep 2005 at 5:48, John Bawden wrote:
> You can test the "multi reel" store to disc as
> follows:
> FILE STD1=STORE1.BACKUP.SYS;DEV=DISC
> FILE STD2=STORE2.BACKUP.SYS;DEV=DISC
>
> STORE some small file set;;RESTORESET=(*STD1),(*STD2);
> COMPRESS=HIGH (or LOW, your choice);SHOW
>
> This will create two disc files or you can create more if you
> choose, I have used up to 5 for large databases.
Thank you for your suggestion. Evidently I am not making myself
clear on what I wish to do. What I wish to test is the mix of MPE
and HFS namespace files that STORE creates when it exceeds the
maximum file size for a single reel, whether or not this
eventuality is anticipated. The method proposed above retains all
files in the MPE namespace, which is fine so long as one is aware
that the incremental step point has not been reached with the
underlying data. My question for you is what happens if one
actually requires three reels when only providing for the two
anticipated? Does one then end up with STORE1.GROUP.ACCT,
STORE2.GROUP.ACCT, and \ACCT\GROUP\STORE1.2 or is it
\ACCT\GROUP\STORE2.2? How do you test and find out what happens
absent 9 Gb of data to store and as much surplus disc capacity?
How do you test and verify your recovery procedures? How much time
does this testing take with 9Gb of data moving over the network
vice some arbitrary lesser amount?
In our case, an FTP transfer that takes 15 minutes between two
linux boxes on the same switch as the HP3000 takes 2 hours and 38
minutes between the HP3000 and the first Linux host. This is for a
single reel that is not quite yet 4GB (nominal). A second 4 Gb
reel would take another 2.75 hours. This is more than 5.5 hours
one-way (and 11 overall) just to verify behaviour that could be
equally well assured with three 1Kb STD files and take an
insignificant amount of transmission time.
Thus I find it most disconcerting that such an important MPE tool
as STORE appears written in such a fashion as that it declines to
fully implement the elegance and entire functionality of the FILE
command. Had this standard behaviour been adhered to then much of
the time and effort I have expended this past month testing STD
either would have been avoided altogether, or much reduced.
Consider the present difficulties encountered testing mixed
namespace disc reel sets. Had FILE been fully implemented then
this would have been a trivial exercise. As it is one is required
to create enough data to exceed the hardcoded limit. Similarly,
allowing one to set the block size for the STD would have made
moving the resulting STD reels to tape a simple exercise using
TAPECOPY (although I may have found a Linux solution to this
difficulty).
We are headed off the HP3000 over the next few years and I do not
expect that this (in my view) defect in STORE will ever be
addressed. Nonetheless, I find it disturbing that such inconsistent
behaviour with respect to MPE was introduced into such an essential
part of the OS. It can only speak to the infrequency of STD use and
the availability of manual work-arounds such as the proposal above
that the situation persists. It is important to me that I see our
backup and recovery procedures actually provide the expected
behaviour under our operating conditions and are robust enough to
function reliably in the absence of continuous monitoring. The
multi-namespace behaviour of STORE and the presence of hard limits
make this testing much more time consuming and difficult than is
otherwise necessary.
An alternative design approach would not have been so difficult
either. I can easily conceive of STORE STD filesets as being
analogous to IMAGE datasets. It surely would have posed no
hardship to fixing the user specified portion of an STD to six
characters and then have STORE completely manage the data reels
from 00 to 99, yielding 0.4Tb of disc store while keeping the
entire thing comfortably inside the MPE namespace. I imagine that
there exist some HP3000 users whose STORE STD requirements exceed
this, but I doubt that there are very many that even have this
amount of disc space connected to their HP3000. For those that do
require STDs greater than 400Gb then the logical thing is to
require multiple STORE reel-sets specified as above, yielding
essentially an infinite amount of STD entirely within the MPE
namespace.
Posix and the HFS was, in my opinion, a great addition to MPE but I
wish that programmers of MPE FOS utilities had avoided visibly
mixing the the two namespaces. An HP3000 user should have retained
the option of staying "pure" MPE without concerning themselves with
default behaviour of FOS utilities that involve the HFS namespace.
Sigh,
Jim
--
*** e-mail is not a secure channel ***
mailto:byrnejb.<token>@harte-lyne.ca
James B. Byrne Harte & Lyne Limited
vox: +1 905 561 1241 9 Brockley Drive
fax: +1 905 561 0757 Hamilton, Ontario
<token> = hal Canada L8E 3C3
* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *
|