Re: 2.6.2-rc2 Interactivity problems with SMP + HT

From: Bill Davidsen
Date: Sun Feb 01 2004 - 16:01:15 EST


Catalin BOIE wrote:
On Thu, 29 Jan 2004, Robert M. Hyatt wrote:



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.


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