DMA vs PIO speeds...

[Guennadi V. Liakhovetski]

Hello everybody

With the help from Andre Hedrick and others and by a horrible dirty hack
in ide-dma.c I managed to enable DMA for a weird (rare?) revision of
Triton chipset in 2.4.1-pre8 kernel (last kernel, when DMA worked for this
chipset was 2.0.39). So, now I've got PIO4 (max throughput 16.6 
MB/sec) and DMA mword2 (also 16.6 MB/sec). Here goes the weird thing:
hdparm -t /dev/hda
with PIO4    - 4.1 MB/sec (4.8 with 2.2.x kernels)
with DMA mw2 - 3.1 MB/sec

I understand I won't be able to change this, but a purely theoretical
question - why? And, can there be (extremely rare on a standalone home
machine) occasions, say, when you need both the CPU and hd IO, when DMA
3.1 would outperform PIO 4.1 (or even 4.8)? I am going to try bonnie
too...

Thanks
Guennadi
___

Dr. Guennadi V. Liakhovetski
Department of Applied Mathematics
University of Sheffield, U.K.
email: [e-mail]