Subject: Re: presto support?
To: Matthew Jacob <email@example.com>
From: Simon Burge <firstname.lastname@example.org>
Date: 02/12/1998 10:50:37
On Wed, 11 Feb 1998 14:36:34 -0800 Matthew Jacob wrote:
> >Right. And the effect isn't the same - in one case, you're getting
> >speed and reliability. In the other case, you're only getting speed.
> >BTW, if the PrestoServe support was implemented the right way, it
> >would speed up local as well as remote disk access.
> Yes, this is true. But also note that PrestoServe memory cards
> are slave only devices. Assuming a TurboChannel or SBus or PCI
> memory card you're stashing data in you pay for not only the
> CPU cycles of moving the data, but moving it in a really undesirable
> manner (across several bus bridges, little help from a cache
> prefetch, and also providing massive interference for other I/O
> that may be occurring). The tradeoffs are not always obvious.
On the DECstation I beleive it sits on "normal" memory space. Certainly
the PrestoServe SIMMs live in a normal memory slot. FWIW, I've got a
little dbx script for Ultrix that turns a piece of normal memory into a
prestoserve buffer (obviously _not_ battery backed).