The longer since the last update of a member, the less likely it is to
be replaced.  Store by last update.  Copying to a new PDS is
alphabetical, and makes the first few compresses particulaly
inefficient (lots of members to be moved).

On Thu, Feb 14, 2013 at 8:37 AM, John Gilmore <[email protected]> wrote:
> Edward Jaffe wrote:
>
> | But would that not require "active" data movement/cleanup
> | whenever a block is freed?
>
> and of course it would.  How much of this would be required appears to
> depend upon what kind of member-replacement process is going on.
>
> There is some evidence---my sample is from six shops but for only 113
> PDSEs---that an 80-20-like process is frequent, i.e., that 80 percent
> of the replacement activity is with only  20 per cent of the members.
> If this is common then keeping a list with counts of the last n
> replacement operations weighted for duplications would make it
> possible to use a very small amount of this sort of thing to great
> advantage.
>
> Shane's point merits comment too.  PDSEs were radically
> under-instrumented at the beginning and still are; for this reason
> discussion of their deficiencies quickly degenerates into competitive
> anecdotage.  One of the largest contributions IBM could make just now
> would be to greatly improve optional SMF-based instrumentation for
> PDSEs,
>
> We all understood the deficiencies of PDSs, and PDSEs were a laudable
> initiative, but they were also a  half-hearted or, perhaps better,
> underbudgeted one.
>
> John Gilmore, Ashland, MA 01721 - USA

-- 
Mike A Schwab, Springfield IL USA
Where do Forest Rangers go to get away from it all?

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to