Jerry Pournelle gets his necktie caught in the wringer a lot
because he doesn't fundamentally know what the heck he's
doing, I've concluded.
For instance, in one column some years ago
I recall he got a PC from NEC or somebody with Win 95 preloaded.
Trouble was, the user name was not Jerry Pournelle, and
he wanted to change it so Jerry Pournelle would come up
as the default document owner, etc.
Well, he couldn't find out how to do that without editing
the registry, and he had heard that it was dangerous to
use the registry editor, so that was out. Ever resourceful,
he turned to one of his HEX editors, and edited the registry
in binary form, found all the instances of whatever the
heck the username was and changed them to Jerry Pournelle.
Worked like a charm. No dangerous REGEDIT for him. Nossir.
Man oh man, I'd rather go bear hunting with pea shooter
than use a hex editor on the registry. Gimme a break.
In this column, what he's seeing is exactly the thrashing
that I have referred to when NT runs low on virtual memory.
Give it long enough, and NT always sorts it out, even if
a program actually has a memory leak. Better you look
in the task manager, find out who owns all that memory,
and shoot the right fish. In fact, that's the only way
to fix it, really. What Pournelle needs is a bigger
page file and better behaved apps.
I don't know for sure, but what I suspect is that memturbo
simply speeds up NT's garbage collection a bit, which, under
extreme circumstances, e.g., thrashing, could conceivably
make a marginal difference. And maybe not. Anyhow,
if a process has a memory
leak, no way memturbo can free memory owned by the leaky
process. If the process had deallocated the memory,
NT could free it. I may have reservations about NT's
paging algorithm, but it's memory management is reputed
to be top notch, and I've found that to be true in practice.
Heck, I've run NT for WEEKS with many many apps open in
64MB. Occasionally I'll scan a bunch of pages and
bring NT to it's knees kicking the page file, but
it always sorts it out sooner or later. In other words,
this is a crock.
>>I'm tempted to check out MemTurbo in particular...
I have read several reviews of various similar sounding
products, including memturbo I'm at least 90% certain,
and none have had more than minimal help, if that.
However, some of them badly destabilize the OS.
Don't remember details, but that's what I've seen, FWIW.
The way Pournelle describes it, memturbo sounds better
than sliced bread, a red flag all by itself IMO. What
does he mean "defrag memory"? That's what page and segment
map tables are for. Maybe if it gets bad enough. I don't
know how Intel memory mapping hardware works, but the
POINT of page mapping is to remove the impact of memory
fragmentation. That's what paging is all about. Segmentation
is different, but doesn't Intel page the segments, for goodness
sake? Anybody know about the Intel page/segment mapping
hardware? Is it a throwback. We learned better than
that over 20 years ago (heck. over 30 years ago). Maybe
I'm to old and things have improved since then <g>.
Spots |
