What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

[Michael B Allen]

I'm running Red Hat 7.3 with their stock 2.4.18-3 kernel on an IBM
T30. Once every few hours X locks up for 5-10 seconds while the disk
grinds. If I type in an Xterm the characters are not echoed until the
disk grinding stops. Then they all come out in a bunch and life is back
to normal.

I asked about this on kernelnewbies but the only response was something
regarding some kind of change to the 'elevator code' but they didn't
know of a solution.

I would like very much for this behavior to go away as it is extremely
annoying. If there is a patch please let me know where I can get it.

Thanks,
Mike

-- 
A  program should be written to model the concepts of the task it
performs rather than the physical world or a process because this
maximizes  the  potential  for it to be applied to tasks that are
conceptually  similar and, more important, to tasks that have not
yet been conceived. 

Re: What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

[Marc-Christian Petersen]

On Tuesday 22 April 2003 09:48, Michael B Allen wrote:

Hi Michael,

> I'm running Red Hat 7.3 with their stock 2.4.18-3 kernel on an IBM
> T30. Once every few hours X locks up for 5-10 seconds while the disk
> grinds. If I type in an Xterm the characters are not echoed until the
> disk grinding stops. Then they all come out in a bunch and life is back
> to normal.
> I asked about this on kernelnewbies but the only response was something
> regarding some kind of change to the 'elevator code' but they didn't
> know of a solution.
The problem is, since 2.4.19-pre5 the elevator has changed to give more 
throughput but also introduce pauses and even stops while disk i/o.

Redhat's 2.4.18 is not 2.4.18 but something 2.4.19ish(-pre|-pre-ac or so).

> I would like very much for this behavior to go away as it is extremely
> annoying. If there is a patch please let me know where I can get it.
There are some hacks. One by Andrea Arcangeli, one by Neil Schemenauer and one 
by Con Kolivas and me. Search the archives please (lowlat elevator/io 
scheduler)

ciao, Marc

Re: What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

[Michael B Allen]

> On Tuesday 22 April 2003 09:48, Michael B Allen wrote:
>
> Hi Michael,
>
>> I'm running Red Hat 7.3 with their stock 2.4.18-3 kernel on an IBM
>> T30. Once every few hours X locks up for 5-10 seconds while the disk
>> grinds. If I type in an Xterm the characters are not echoed until the
<snip>
>> I would like very much for this behavior to go away as it is extremely
>> annoying. If there is a patch please let me know where I can get it.
> There are some hacks. One by Andrea Arcangeli, one by Neil Schemenauer and
> one
> by Con Kolivas and me. Search the archives please (lowlat elevator/io
> scheduler)

Ok, I searched a little using the Googler at indiana.edu's archives but
nothing jumped up and bit me. I'm not too excited about applying a patch
snarfed out of an e-mail anywat. I'm surprised no one else has not
complained about this enough to the point where you guys don't have a
canned answer with a link. Is this problem not considered important?

Does anyone know which RH patch in the 2.4.18-10 RPM adds this elevator
throughput "improvement"? What identifiers would such a patch have in it?

Thanks,
Mike

PS: Why are there only "hacks"? Is this not considered important?

Re: What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

[jw schultz]

On Tue, Apr 22, 2003 at 05:35:02AM -0400, Michael B Allen wrote:
> > On Tuesday 22 April 2003 09:48, Michael B Allen wrote:
> >
> > Hi Michael,
> >
> >> I'm running Red Hat 7.3 with their stock 2.4.18-3 kernel on an IBM
> >> T30. Once every few hours X locks up for 5-10 seconds while the disk
> >> grinds. If I type in an Xterm the characters are not echoed until the
> <snip>
> >> I would like very much for this behavior to go away as it is extremely
> >> annoying. If there is a patch please let me know where I can get it.
> > There are some hacks. One by Andrea Arcangeli, one by Neil Schemenauer and
> > one
> > by Con Kolivas and me. Search the archives please (lowlat elevator/io
> > scheduler)
> 
> Ok, I searched a little using the Googler at indiana.edu's archives but
> nothing jumped up and bit me. I'm not too excited about applying a patch
> snarfed out of an e-mail anywat.

That's how Linus, Marcello and everyone else gets them.

> I'm surprised no one else has not
> complained about this enough to the point where you guys don't have a
> canned answer with a link. Is this problem not considered important?
> 
> Does anyone know which RH patch in the 2.4.18-10 RPM adds this elevator
> throughput "improvement"? What identifiers would such a patch have in it?
> 
> Thanks,
> Mike
> 
> PS: Why are there only "hacks"? Is this not considered important?

It is up to Red Hat to $upply fixes for their kernels.


-- 
________________________________________________________________
	J.W. Schultz            Pegasystems Technologies
	email address:		jw@pegasys.ws

		Remember Cernan and Schmitt

Re: What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

[Marc-Christian Petersen]

[-- Attachment #1: Type: text/plain, Size: 1523 bytes --]

On Tuesday 22 April 2003 11:35, Michael B Allen wrote:

Hi Michael,

> Ok, I searched a little using the Googler at indiana.edu's archives but
> nothing jumped up and bit me. I'm not too excited about applying a patch
Then you are not able to search archives.

http://marc.theaimsgroup.com/?l=linux-kernel&m=105092498721316&w=2
http://marc.theaimsgroup.com/?l=linux-kernel&m=105060066815681&w=2
http://marc.theaimsgroup.com/?l=linux-kernel&m=103835185429978&w=2

The 3rd url has a config option. Same attached w/o config option.

> snarfed out of an e-mail anywat. I'm surprised no one else has not
> complained about this enough to the point where you guys don't have a
> canned answer with a link. Is this problem not considered important?
there were some complaints. Anyway, if it's important or not depends on the 
point of view. For me, it is _mega_ important, for mainline 2.4, I don't 
think it is important ;)

> Does anyone know which RH patch in the 2.4.18-10 RPM adds this elevator
> throughput "improvement"? What identifiers would such a patch have in it?
Look for a patch that adds "2.4.19-pre*".

> PS: Why are there only "hacks"? Is this not considered important?
Well, the attached one is really a hack. It drops throughput ~20-30 % but gets 
rid of all the annyoing pauses/stops while disk i/o. The 2 other approaches 
are no real hacks though they also drop throughput.

You can also try "elvtune -r 0 -w 8192 /dev/bla" and see if it makes any 
difference if you don't want to patch anything.

ciao, Marc

[-- Attachment #2: lowlat-elevator-hack.patch --]
[-- Type: text/x-diff, Size: 950 bytes --]

--- old/drivers/block/ll_rw_blk.c	2002-12-18 01:03:50.000000000 +0100
+++ new/drivers/block/ll_rw_blk.c	2003-04-22 11:49:36.000000000 +0200
@@ -432,9 +432,7 @@ static void blk_init_free_list(request_q
 
 	si_meminfo(&si);
 	megs = si.totalram >> (20 - PAGE_SHIFT);
-	nr_requests = 128;
-	if (megs < 32)
-		nr_requests /= 2;
+	nr_requests = 4;
 	blk_grow_request_list(q, nr_requests);
 
 	init_waitqueue_head(&q->wait_for_requests[0]);
--- old/include/linux/elevator.h	2002-12-18 01:03:59.000000000 +0100
+++ new/include/linux/elevator.h	2003-04-22 11:49:58.000000000 +0200
@@ -93,8 +93,8 @@ static inline int elevator_request_laten
 
 #define ELEVATOR_LINUS							\
 ((elevator_t) {								\
-	2048,				/* read passovers */		\
-	8192,				/* write passovers */		\
+	0,				/* read passovers */		\
+	0,				/* write passovers */		\
 									\
 	elevator_linus_merge,		/* elevator_merge_fn */		\
 	elevator_linus_merge_req,	/* elevator_merge_req_fn */	\

Re: What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

[Alan Cox]

On Maw, 2003-04-22 at 10:56, jw schultz wrote:
> > >> I'm running Red Hat 7.3 with their stock 2.4.18-3 kernel on an IBM
> > >> T30. Once every few hours X locks up for 5-10 seconds while the disk

> It is up to Red Hat to $upply fixes for their kernels.

2.4.18-3 is old old old stuff. There are multiple errata for it.

Re: What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

[Alan Cox]

On Maw, 2003-04-22 at 08:48, Michael B Allen wrote:
> I asked about this on kernelnewbies but the only response was something
> regarding some kind of change to the 'elevator code' but they didn't
> know of a solution.

Its actually probably not that
> 
> I would like very much for this behavior to go away as it is extremely
> annoying. If there is a patch please let me know where I can get it.

See the Red Hat 8 errata. 

Re: What's the deal McNeil? Bad interactive behavior in X w/ RH's 2.4.18

[J Sloan]

Michael B Allen wrote:

>Ok, I searched a little using the Googler at indiana.edu's archives but
>nothing jumped up and bit me. I'm not too excited about applying a patch
>snarfed out of an e-mail anywat. I'm surprised no one else has not
>complained about this enough to the point where you guys don't have a
>canned answer with a link. Is this problem not considered important?
>
Yes, the link is www.redhat.com/errata

But yes, for those who like to roll their
own, Con's responsiveness patchset is
farily well regarded and trusted - you
can grab it from kernel.kolivas.org -

I've had good results with just the 001,
002 and 003 patches on stock 2.4.20 -

>
>Does anyone know which RH patch in the 2.4.18-10 RPM adds this elevator
>throughput "improvement"? What identifiers would such a patch have in it?
>  
>
2.4.18-10?

You want 2.4.18-27.7x - in other words, the
current redhat kernel for your release.

The patches are all in the kernel srpm for
you to view and analyze...

BTW If you really want low latency you'll
love 2.6

Joe