stuart@grapevine.uucp (Stuart J. McIntyre) (08/22/89)
Situation: A friend has the following setup: A2000, 1.3 ROM, 1 Mb Agnus, A2090A, Quantum Q280 (80Mb) 2 internal floppies, and a Ronin Hurricane 68030/68881 board with 4 Mb 32bit Ronin memory. (no 16bit Fast RAM) Commentary/Questions: The first partition on the HD is cyl 0-3, Old file system, followed by four FFS partitions; 6.5 Mb, 30Mb, 30Mb, 11Mb. With the ronin CPU/memory OFF the machine works great. With maxtransfer set to 262144 on the FFS partitions we were getting 508K/sec. But!!! in order to use the 68030 we had to add "Mask = 0xfffff" to the mountlist for each of the FFS partitions. This dropped diskperf down to 80k/sec!! What should the Maxtransfer be to avoid all r/w errors? I remember people talking about this but at the time wasn't interested. What does the "Mask" do? Is its purpose to avoid DMA'ing into the Ronin memory? (Notes: Without the above Mask entry in the mountlist, the machine guru'ed at any attempt to run Hurricaneconfig or Hurricaneconfig -a -r followed by addmem32. With the Mask, Hurricaneconfig -r was still required to avoid the guru, (so now we have a 20 MHZ Amiga with a quantum running at 80k/sec and 4Mb 32 bit ram but we can't remap kickstart into the fast ram without crashing.) Will a trump card (or Kronos card) solve the DMA/fast RAM problems? Thanks for any hints/insights! -Stuart ------------------------------------------------------------------- Stuart McIntyre stuart@chrome.sun.com Disclaimer: My views do not represent the views of Sun Microsystems.
daveh@cbmvax.UUCP (Dave Haynie) (08/23/89)
in article <34149@grapevine.uucp>, stuart@grapevine.uucp (Stuart J. McIntyre) says: > Keywords: 2090 2000 1.3 Ronin 68030 > With the ronin CPU/memory OFF the machine works great. With maxtransfer > set to 262144 on the FFS partitions we were getting 508K/sec. But!!! > in order to use the 68030 we had to add "Mask = 0xfffff" to the mountlist > for each of the FFS partitions. This dropped diskperf down to 80k/sec!! Yup. Look carefully at that mask value. It has the effect of restricting all DMA to Chip memory. The result of this is that you're fast DMA disk is now running about as slow as possible in such a system. The only way you can speed it up at all would be to DMA into Fast memory and then CPU copy to Ronin RAM, but that's not going to be all that much faster. This is a two-way problem. Obviously a DMA controller can't get to RAM that's not in it's address space. Only, the Ronin memory normally sits in autoconfig space, though it doesn't autoconfig, and it's not DMA-able. That's pure evil, there aren't many other rule this Ronin RAM could have broken. There's supposed to be a way to move it up out of the autoconfig space -- I'd recommend that, as that's where such memory belongs, if you have it -- out of the 24 bit address space. Then you can benefit as much as possible from expansion bus Fast memory, and set your mask to 0xfffffe to keep the DMA within the 24 bit space. > required to avoid the guru, (so now we have a 20 MHZ Amiga with a > quantum running at 80k/sec and 4Mb 32 bit ram but we can't remap > kickstart into the fast ram without crashing.) I suspect you have a 14.3 Mhz Amiga, perhaps with a 20 or 25MHz math chip, unless this is a new version of the Ronin card. I hope not -- I'd like to think they wouldn't introduce the same flaws twice. They were the first 3rd party 68020 design to at least get the basic bus interface stuff 100%, at least as much as I've tested the thing (played with one for 2 weeks); why stop at halfway right? > Will a trump card (or Kronos card) solve the DMA/fast RAM problems? Any non-DMA card will work faster in such a setup; C Ltd., IVS, or GVP are ones I know of. They still won't be as fast as DMA could be with an A2620 or A2630, but they'll work FAR better than your current setup. > -Stuart -- Dave Haynie Commodore-Amiga (Systems Engineering) "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: hazy BIX: hazy We have no choice. We are, after all, professionals.