From: Drew Eckhardt (drew@ophelia.cs.colorado.edu)
Date: 05/15/92


From: drew@ophelia.cs.colorado.edu (Drew Eckhardt)
Subject: Re: Abysmal disk performance
Date: Fri, 15 May 1992 22:27:22 GMT

In article <1992May14.215250.3430@news.cso.uiuc.edu> ddd42961@uxa.cso.uiuc.edu (Daniel D Deavours) writes:
>
>Never programmed a DMA before, but if I'm correct, the advantage may not be
>so much speed in data transfer, but the ability to do disk i/o in the back-
>ground. It's annoying, especially on my slow hd/controller for everything to
>wait until the hard drive is done (especially when syncing.)

1. If you're talking about a "bus master", it can steal cycles and DMA
        can happen in the background.

2. If you're talking about DMA initiated by the motherboard,
        it cannot happen in the background (unless the chipset
        allows snooping from the cache), and the advantage is
        speed.

sync() does not exit until the buffercache is flushed, for obvious reasons.