>SNIP
>>
>>From an OS-design standpoint I don't think this has to be so - virtual-memory subsystems have been around for a long time and are well understood, but if that's what you're seeing in the real world under heavy loads I'm not going to argue. From time to time I've encountered articles in the MSKB outlining server OS issues under heavy loads - IIRC most were about NT4, I don't know if they've been addressed in W2K or Server 2003. Sometimes hotfixes address these issues, which can be with specific pieces of hardware. Again, since I don't see those sorts of loads I can't offer any real-world insight.
>>
>SNIP
>
>I just want to comment, Al, that though virtual memory systems HAVE been around for a long time, I would say that the longest 'living' one that I am familiar with - IBM's on /370 (and later) equipment - is radically different than the MS flavour.
>
>IBM's involved, right from the start, a hardware+software solution. In addition IBM's offered options to keep things from paging/swapping as well as some user control over page/swap susceptibility. IBM's also differentiated "paging" from "swapping" and used the concept of the "working set" right from the start.
>Just from direct observation, as well as some reading, it is apparent that MS' techniques vary significantly from IBM's in MOST aspects.
>
>I have the opinion that MS, in the cacheing area, has gone way overboard in trying to apply 'intelligence' to maximize throughput while essentially 'hiding' the operation from users and keeping users away from any control of it. The Redirector only adds additional complexity and the MS credo (which is a good one for the customer) of having all combinations of OS' interoperate further complicates an already sensitive area.

With cache designs there's more than one way to skin a cat. IBM's and MS's designs may be different but that's not to say that one is better than the other or that either is wrong. I don't recall seeing any articles or test results showing that MS's scheme is consistently unreliable, and it gets hammered every day by literally millions of systems. So although it may not be perfect (nothing is), I'd say it has a pretty good track record.

>Personally I think that MS would do themselves a BIG service, and us too, by better explaining the details and, more importantly, by giving users a few choices as regards their own deployment of cacheing (sure, it can be argued we have some now, in the form of registry settings, but these are obscure, an all-or-nothing proposition and ill-explained. No one should have to cross their fingers and toes when they make settings changes and in the case of cacheing settings that decidedly is the case.

Information/details - yes, I've been asking for this for some time. Control? Definitely a two-edged sword. The idea of people who don't know exactly what they're doing, messing with cache settings makes me shudder.