home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.aux
- Path: sparky!uunet!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!cmcl2!panix!alexis
- From: alexis@panix.com (Alexis Rosen)
- Subject: Re: Mac accelerators and A/UX 3.0
- Message-ID: <1992Jul28.210125.23387@panix.com>
- Date: Tue, 28 Jul 1992 21:01:25 GMT
- References: <1992Jul27.213533.4284@noao.edu> <5!mmv7#.stever@netcom.com> <1992Jul28.020627.10142@noao.edu> <aymmx0d.stever@netcom.com>
- Organization: PANIX Public Access Unix, NYC
- Lines: 25
-
- stever@netcom.com (Steve Riggins) writes:
- >Can anyone run tests for us to see if the DMA on a IIfx really helps any at all?
-
- Interestingly enough, this was a real issue. In 2.0.0, the DMA mode was a win
- for some things and a big lose for others. If I recall correctly, this was
- due to the kernel parameter for minimum dma xfer size being set too low. I
- fiddled with it and it got faster.
-
- In 2.0.1, it was fixed somewhat but I didn't test it.
-
- 3.0 was a lot faster than 2.0. The DMA definitely made a perceptible difference
- in throughput. The raw I/O speed hasn't really changed (much?) but the load
- on the rest of the system from disk I/Os has decreased a lot.
-
- The problem is that DMA is only useful in some situations. On the IIfx, the
- disk I/O channel is mismatched to the CPU speed- it's too slow. DMA can't
- help you there, and in fact, while it's sometimes an improvement, it also
- highlights that problem. The Quadras are _much_ better machines from that
- standpoint. Especialy the 9x0.
-
- --
- Alexis Rosen Owner/Sysadmin,
- PANIX Public Access Unix & Internet, NYC.
- alexis@panix.com
- {uupsi,cmcl2}!panix!alexis
-