Re: 2.6.2-rc2 Interactivity problems with SMP + HT

From: Catalin BOIE
Date: Mon Feb 02 2004 - 08:06:39 EST


> >>It might be some IDE disk I/O that results from flushing buffers or
> >>whatever. I don't see this on my SCSI boxes, but I have seen an IDE
> >>box get sluggish at times due to I/O.
> >
> >
> > It is possible.
> > vmstat shows a lot of writes when this happen.
> > Seems that even reads hangs.
> > I remember tat I was in pine and I tried to save a small file (under 1k)
> > and it took 5-7 seconds to do it.
> >
> >
> >>On Thu, 29 Jan 2004, Nick Piggin
> >>wrote:
> >>
> >>
> >>>
> >>>Catalin BOIE wrote:
> >>>
> >>>
> >>>>Hello!
> >>>>
> >>>>First, thank you very much for the effort you put for Linux!
> >>>>
> >>>>I have a Intel motherboard with SATA (2 Maxtor disks).
> >>>>CPUs: 2 x 2.4GHz PIV HT = 4 processors (2 virtual)
> >>>>1 GB RAM.
> >>>>
> >>>>Load: postgresql and apache. Very low load (3-4 clients).
> >>>>
> >>>>RAID: Yes, soft RAID1 between the 2 disks.
> >>>>
> >>>>I have times when the console freeze for 3-4 seconds!
> >>>>2.6.0-test11 had the same problem (maybe longer times).
> >>>>2.6.1-rc2 worked good in this respect but crashed after 2 days. :(
> >>>>2.6.2-rc2 is back with the delay.
> >>>>
> >>>>Do you know why this can happen?
> >>>>
> >>>>
> >>>
> >>>There haven't been many scheduler changes there recently so
> >>>maybe its something else.
> >>>
> >>>But you could try the latest -mm kernels. They have some
> >>>Hyperthreading work in them (you need to enable CONFIG_SCHED_SMT).
>
> Another possible improvement might come from the recent ide changes
> posted here for evaluation. There was a change to not block all devices
> or even all devices on a cable if one device delayed. I believe it was
> if a command didn't finish as intended, but I don't have it handy since
> I sent it off for testing tomorrow on a system which may have that problem.
>
> There's a good bit of work on hangs happening right now, so you could
> wait or dig up the patches and try them. I think they were by Davin
> McCall if memory serves.

Thank you very much.

>
>
> --
> bill davidsen <davidsen@xxxxxxx>
> CTO TMR Associates, Inc
> Doing interesting things with small computers since 1979
>

---
Catalin(ux) BOIE
catab@xxxxxxxxxxxxxxx
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/