[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Does Load-to-Paging Migration really help paging performance?

Like most simple questions, the answers to yours are not simple.

    Date: Sun, 20 Dec 87 16:08:12 PST
    From: luria@renoir.Berkeley.EDU (Marc Luria)

    Does Load-to-Paging Migration really help paging performance?

Yes in some benchmarks, no in others.  If your benchmark is spending
time referencing dynamic objects, or if it is making more-or-less random
references to all objects, then Load-to-Paging Migration won't help you
much.  However, if you are referencing localized groups of objects which
existed in your world load, where some are modified and others aren't,
then you will spend more time paging as the disk seeks between the
fragmented pages on the disk.

For most applications enabling Load-to-Paging Migration has no great
effect.  If you're really interested, benchmark your application both
ways, then decide.  

    Is it worth having to allocate extra paging for that 80% that doesn't ever
    get changed?

That's a trade-off which you'll have to evaluate.

		  (p. 148 of Vol. 0) How much of that 80% actually ever gets read?

Depends what you do.  If you don't execute FED, you'll never reference
FED functions and datastructures.

Finally, the 80% in the documentation is an estimate.  It could vary
(and does vary) widely.

I recommend leaving Load-to-Paging Migration alone.