It seems Peter Wemm wrote:
> > There is a HW problem with the HPT366 and "sufficiently" fast disks,
> > I've been working with highpoint on that, and so far there is no
> > known solution, other than slowing down the interface speed, ie
> > setting it back to ata33 would make it work with my DTLA drives...
>
> Hmm. I was doing that for around 8 months but still got occasional lockups
> when 6 disks were active over three controllers. I suspected that the
> devices were not adequately robust when dealing with multiple FIFO
> overflows when the PCI bus saturated, as the combined raw disk bandwidth
> plus ethernet exceeded the PCI bus maximum speed, let alone burst speeds
> from disk cache hits. Anyway, using the promise cards "solved" the problem
> but at the cost of significant slowdown.
>
> eg: the HPT366 was two pci devices with two busmaster engines. I could run
> two drives at 19 or 20 MByte/sec in parallel with one drive on each channel.
> The per-drive speed was the same for single or dual disks.
>
> With the promise cards though, a single disk maxed at ~18MByte/sec and two
> disks slowed down to ~14-15MByte/sec each. It seemed as though the promise
> was overcommitted to running two channels in parallel. The PIIX4 has the same
> problem but is far far worse (ie: slows down to 10MByte/sec each).
>
> Out of curiosity, do you know of similar problems for the HPT370?
No, and the Promises that I have (33, 66 & 100's) doesn't suffer from
the problem you descipe at all, I only see slowdowns when I reach a
combined throughbout of ~110MB/sec thus saturating the PCI bus...
-Søren
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
