I`ve recently spotted that a solo1 pci soundcard generates
16000+ interrupts/second with esd started idling.
What strikes me as odd is that my old good isa sb16 vibra
does only 600+ interrupts.
Also i`ve had some strange experience of this interrupt load
affecting the disk io.
The situation was the next: i`ve had a large transfer to a samba
boxen with this poor solo1 soundcard. I would like to emphasize that i
turn off the drive write-caching ability in order to gain reliability,
and that with write cacheing the following issues are hardly noticeable.
So yes, i was copying large amount of data thru the 10mbit rtl8029s
to the samba box. What i`ve tested are two situations: with esd on ==
effective 16,5k interrupts/second, and with esd off, and thus no interrupts
generated by the solo1 card.
What i`ve noticed quite disappointed me: with esd turned on, the disk
writeout activity (well, the seek activity) was _much_ higher.
Ie i`ve stopped esd, the seekage dropped down, i`ve started esd -
seekage goes back.
So we basically have here two strange issues:
1. Ess Solo-1 driver generates awful lots of interrupts (16k/s)
2. Heavy interrupt load forces the elevator to make bad
decisions
Info: linux-2.4.17-rmap12e, p2-300 128M RAM, rtl8029 NIC, Ess Solo-1 SB
regards, Samium Gromoff
> I`ve recently spotted that a solo1 pci soundcard generates
> 16000+ interrupts/second with esd started idling.
Thats an esd bug. ESD tries to use ridiculously small fragment sizes
" Alan Cox wrote:"
>
> > I`ve recently spotted that a solo1 pci soundcard generates
> > 16000+ interrupts/second with esd started idling.
>
> Thats an esd bug. ESD tries to use ridiculously small fragment sizes
>
Wait, wait, but my ISA Vibra 16 generates 20+ times less interrupts, with the
_same_ esd!
On Tue, 19 Feb 2002, Samium Gromoff wrote:
> Wait, wait, but my ISA Vibra 16 generates 20+ times less interrupts, with the
> _same_ esd!
Might have something to do with the way the driver handles things, sb
being one of the better ones, i'm sure this was trawled around a couple
months back Re: interrupt storm and emu10k1
Cheers,
Zwane Mwaikambo
> > Thats an esd bug. ESD tries to use ridiculously small fragment sizes
> >
> Wait, wait, but my ISA Vibra 16 generates 20+ times less interrupts, with the
> _same_ esd!
Yes. It has diff fragment limits
At some point in the past, Samium Gromoff wrote:
>>> I`ve recently spotted that a solo1 pci soundcard generates
>>> 16000+ interrupts/second with esd started idling.
" Alan Cox wrote:"
>> Thats an esd bug. ESD tries to use ridiculously small fragment sizes
I'm not sure what it is, but I traded in esd, the sound card, and the
scheduler upon gaining greater insight into this.
On Tue, Feb 19, 2002 at 04:44:28PM +0300, Samium Gromoff wrote:
> Wait, wait, but my ISA Vibra 16 generates 20+ times less interrupts, with the
> _same_ esd!
A lot of it is the hardware. I haven't looked very hard at the drivers,
but my time/money tradeoff seemed to scream "just get new hardware you
don't have time to look at the code down there". Maybe someone else does...
Any volunteers? I'd at least be willing to run tests. I've seen this too.
Cheers,
Bill
" Alan Cox wrote:"
>
> > > Thats an esd bug. ESD tries to use ridiculously small fragment sizes
> > >
> > Wait, wait, but my ISA Vibra 16 generates 20+ times less interrupts, with the
> > _same_ esd!
>
> Yes. It has diff fragment limits
>
So the point is we should fix esd, not the solo-1 driver, i presume?
(esd_fixed -> irq_load_fixed -> disk_io_is_back)... sounds ok
EOT
regards, Samium Gromoff
On 19-Feb 05:12, Samium Gromoff wrote:
> " Alan Cox wrote:"
> >
> > > > Thats an esd bug. ESD tries to use ridiculously small fragment sizes
> > > >
> > > Wait, wait, but my ISA Vibra 16 generates 20+ times less interrupts, with the
> > > _same_ esd!
> >
> > Yes. It has diff fragment limits
> >
> So the point is we should fix esd, not the solo-1 driver, i presume?
> (esd_fixed -> irq_load_fixed -> disk_io_is_back)... sounds ok
>
> EOT
>
> regards, Samium Gromoff
It would be EOT, but there don't seem to _be_ any esd developers. I'd take a
look but my C skills max out at about "hello world", and there wouldn't be
anyone to make a new release (google couldn't find a esd homepage at least).
Maybe the gnome/ximian folks are taking bug reports/fixes in their packages.
Who knows? This has come up several times with nary a peep from a esd
developer.
Thomas
Samium Gromoff wrote:
> " Alan Cox wrote:"
> > Yes. It has diff fragment limits
> So the point is we should fix esd, not the solo-1 driver, i presume?
> (esd_fixed -> irq_load_fixed -> disk_io_is_back)... sounds ok
Note that solo1 has differnet limits depending whether the program (esd)
runs as root or not. root may shoot him/herself into the foot...
Tom