> the Library would spend (4*600)/60 = 40 hours just mounting tapes!
..
Quite right, but only valid if
we consider RESTORE of ALL nodes at once
and the library has only ONE drive.
On the other side, mixing more nodes
on one tape by limiting number of
collocating volumes will add extra search times
to your restore.
For example, if you were limiting
tapes count to 1/3 number of nodes,
resulting tape volume could contain - let us
say, after a week - data from:
node1,node2,node3,node1,node2,node1,node3,node1
Thus restoring NODE1 will add 3 extra searches (first one is over
,node2,node3, second one over ,node2, third one over ,node3,)
which counts for maybe 2 minutes when using DLT.
After a year (50 weeks) this tape contains much longer node mix,
causing maybe 2*50 = about 1.5 hours extra search times.
You have 600 *1/3 tapes = 200 tapes * 1.5 hours = 300 hours penalty
Which is much worse than the above tape mount penalty.
Surely , this is likely to be worst case,
and it depends strongly on your data,
on your policies etc.
Anyway, your example showed very nice how important
the right configuration and
the type of hardware used are,
and that the DLT´s so loved for their
speed are not best for all purposes
(as the case with virtually all technologies is).
Speaking about DLT´s you have to add
relatively slow search times
and disastreously slow response when
trying to read tapes with read errors
I already had an AUDIT ran 2 days
on DLT 4000.
My conclusion: I should better have bought
either 3575 or AIT.
Juraj Salak
Linz, Austria
|