* 2.4.7-ac4 disk thrashing
@ 2001-08-04 15:38 Tom Vier
2001-08-04 23:04 ` Matthew Gardiner
0 siblings, 1 reply; 9+ messages in thread
From: Tom Vier @ 2001-08-04 15:38 UTC (permalink / raw)
To: linux-kernel
switching from 2.4.7-ac3 to -ac4, disk access seems to be much more
synchronis. running a ./configure script causes all kinds of trashing, as
does installing .debs. i'm using reiserfs on top of software raid 0 on an
alpha.
--
Tom Vier <tmv5@home.com>
DSA Key id 0x27371A2C
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: 2.4.7-ac4 disk thrashing
2001-08-04 15:38 2.4.7-ac4 disk thrashing Tom Vier
@ 2001-08-04 23:04 ` Matthew Gardiner
2001-08-05 0:55 ` Tom Vier
2001-08-05 1:09 ` ip_conntrack bigger, better and yellow ;) !!! Mircea Ciocan
0 siblings, 2 replies; 9+ messages in thread
From: Matthew Gardiner @ 2001-08-04 23:04 UTC (permalink / raw)
To: Tom Vier; +Cc: linux-kernel
Tom Vier wrote:
>switching from 2.4.7-ac3 to -ac4, disk access seems to be much more
>synchronis. running a ./configure script causes all kinds of trashing, as
>does installing .debs. i'm using reiserfs on top of software raid 0 on an
>alpha.
>
Apparently, in ac5 (which I am running), there was a bug on non-x86
cpu's using reiserfs. Download and install the new patch and try.
Matthew Gardiner
_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: 2.4.7-ac4 disk thrashing
2001-08-04 23:04 ` Matthew Gardiner
@ 2001-08-05 0:55 ` Tom Vier
2001-08-05 1:09 ` ip_conntrack bigger, better and yellow ;) !!! Mircea Ciocan
1 sibling, 0 replies; 9+ messages in thread
From: Tom Vier @ 2001-08-05 0:55 UTC (permalink / raw)
To: Matthew Gardiner; +Cc: linux-kernel
On Sun, Aug 05, 2001 at 11:04:59AM +1200, Matthew Gardiner wrote:
> Tom Vier wrote:
> >switching from 2.4.7-ac3 to -ac4, disk access seems to be much more
> >synchronis. running a ./configure script causes all kinds of trashing, as
> >does installing .debs. i'm using reiserfs on top of software raid 0 on an
> >alpha.
> Apparently, in ac5 (which I am running), there was a bug on non-x86
> cpu's using reiserfs. Download and install the new patch and try.
that's just a signedness fix. i've tried ac5 and it has the same problem as
ac4.
--
Tom Vier <tmv5@home.com>
DSA Key id 0x27371A2C
^ permalink raw reply [flat|nested] 9+ messages in thread
* ip_conntrack bigger, better and yellow ;) !!!
2001-08-04 23:04 ` Matthew Gardiner
2001-08-05 0:55 ` Tom Vier
@ 2001-08-05 1:09 ` Mircea Ciocan
2001-08-05 3:19 ` Sven Koch
1 sibling, 1 reply; 9+ messages in thread
From: Mircea Ciocan @ 2001-08-05 1:09 UTC (permalink / raw)
To: linux-kernel
Ok, the pressure for a netflow like stuff in kernel is big, examining
the the netflow structure at:
http://sgi.rdscv.ro/~ionuts/netflowv5.h
it seem that ip_conntrak has allmost everything that is needed to
emulate Cisco netflow EXCEPT ( a big except :) the information about
data bytes/packets that flow via that connexion and the question is how
could be added with minimum damages, for example at the end of the
existing ip_conntrack structure, so a nice little userspace daemon could
parse the /proc/net/ip_conntrack and generate the damned netflow packets
that everybody seem to want now :( !!!
Me and a couple of friends are ready to cut our teeth in kernel
programing and have available some machines to crash test and debug the
damned thing and while not aiming for inclusion in stable series maybe
it will mature enough for inclusion in 2.6 series or it will remain a
forever unofficial pach :).
While we think that we know enough C and we can dig throu kernel
sources we need some mentoring, kickstart and stuff from you network
gurus to tell us where is the best this code to be inserted ( is your
stack after all ;) and some best design ideeas.
We promise to do the grunt work and keep the "bothering level" at a
minimum :).
So Alexey, DaveM, Alan and the other network gurus, please help us to
put our name somewhere in the CREDITS file ;).
Thank you,
Mircea C.
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: ip_conntrack bigger, better and yellow ;) !!!
2001-08-05 1:09 ` ip_conntrack bigger, better and yellow ;) !!! Mircea Ciocan
@ 2001-08-05 3:19 ` Sven Koch
0 siblings, 0 replies; 9+ messages in thread
From: Sven Koch @ 2001-08-05 3:19 UTC (permalink / raw)
To: Mircea Ciocan; +Cc: linux-kernel
On Sun, 5 Aug 2001, Mircea Ciocan wrote:
> Ok, the pressure for a netflow like stuff in kernel is big, examining
> the the netflow structure at:
>
> http://sgi.rdscv.ro/~ionuts/netflowv5.h
>
> it seem that ip_conntrak has allmost everything that is needed to
> emulate Cisco netflow EXCEPT ( a big except :) the information about
> data bytes/packets that flow via that connexion and the question is how
> could be added with minimum damages, for example at the end of the
> existing ip_conntrack structure, so a nice little userspace daemon could
> parse the /proc/net/ip_conntrack and generate the damned netflow packets
> that everybody seem to want now :( !!!
I think you should get in contact with the netfilter-coreteam about this
and subscribe to the netfilter-devel-mailinglist - see
http://netfilter.samba.org/ for subscribe-information.
c'ya
sven
--
The Internet treats censorship as a routing problem, and routes around it.
(John Gilmore on http://www.cygnus.com/~gnu/)
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: 2.4.7-ac4 disk thrashing
2001-08-08 15:41 ` Daniel Phillips
@ 2001-08-08 16:03 ` Dieter Nützel
0 siblings, 0 replies; 9+ messages in thread
From: Dieter Nützel @ 2001-08-08 16:03 UTC (permalink / raw)
To: Daniel Phillips, Alan Cox
Cc: Linux Kernel List, ReiserFS List, Chris Mason, Nikita Danilov, Tom Vier
Am Mittwoch, 8. August 2001 17:41 schrieb Daniel Phillips:
> On Wednesday 08 August 2001 12:57, Alan Cox wrote:
> > > Could it be that the ReiserFS cleanups in ac4 do harm?
> > > http://marc.theaimsgroup.com/?l=3Dreiserfs&m=3D99683332027428&w=3D2
> >
> > I suspect the use once patch is the more relevant one.
>
> Two things to check:
>
> - Linus found a bug in balance_dirty_state yesterday. Is the
> fix applied?
No, I'll try.
> - The original use-once patch tends to leave a referenced pages
> on the inactive_dirty queue longer, not in itself a problem,
> but can expose other problems. The previously posted patch
> below fixes that, is it applied?
>
> To apply (with use-once already applied):
Yes, it was with -ac9.
But it wasn't much different from ac6/7/8 without it. All nearly "equally
bad". The disk steps like mad compared against 2.4.7-ac1 and ac-3. I can
"hear" it and the whole system "feels" slow.
2.4.7-ac1 + transaction-tracking-2 (Chris) + use-once-pages
(Daniel) + 2.4.7-unlink-truncate-rename-rmdir.dif (Nikita) is the best Linux
I've ever run.
I did several (~10 times) dbench-1.1 (should I retry with dbench-1.2?) and
all gave nearly same results.
ac-1, ac3 + fixes GREAT
ac5, ac6, ac7, ac8, ac9 + fixes BAD
Thanks,
Dieter
> cd /usr/src/your.2.4.7.source.tree
> patch -p0 <this.patch
>
> --- ../2.4.7.clean/mm/filemap.c Sat Aug 4 14:27:16 2001
> +++ ./mm/filemap.c Sat Aug 4 23:41:00 2001
> @@ -979,9 +979,13 @@
>
> static inline void check_used_once (struct page *page)
> {
> - if (!page->age) {
> - page->age = PAGE_AGE_START;
> - ClearPageReferenced(page);
> + if (!PageActive(page)) {
> + if (page->age)
> + activate_page(page);
> + else {
> + page->age = PAGE_AGE_START;
> + ClearPageReferenced(page);
> + }
> }
> }
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: 2.4.7-ac4 disk thrashing
2001-08-08 10:57 ` Alan Cox
@ 2001-08-08 15:41 ` Daniel Phillips
2001-08-08 16:03 ` Dieter Nützel
0 siblings, 1 reply; 9+ messages in thread
From: Daniel Phillips @ 2001-08-08 15:41 UTC (permalink / raw)
To: Alan Cox, Dieter Nützel
Cc: Linux Kernel List, ReiserFS List, Chris Mason, Nikita Danilov,
Daniel Phillips, Tom Vier
On Wednesday 08 August 2001 12:57, Alan Cox wrote:
> > Could it be that the ReiserFS cleanups in ac4 do harm?
> > http://marc.theaimsgroup.com/?l=3Dreiserfs&m=3D99683332027428&w=3D2
>
> I suspect the use once patch is the more relevant one.
Two things to check:
- Linus found a bug in balance_dirty_state yesterday. Is the
fix applied?
- The original use-once patch tends to leave a referenced pages
on the inactive_dirty queue longer, not in itself a problem,
but can expose other problems. The previously posted patch
below fixes that, is it applied?
To apply (with use-once already applied):
cd /usr/src/your.2.4.7.source.tree
patch -p0 <this.patch
--- ../2.4.7.clean/mm/filemap.c Sat Aug 4 14:27:16 2001
+++ ./mm/filemap.c Sat Aug 4 23:41:00 2001
@@ -979,9 +979,13 @@
static inline void check_used_once (struct page *page)
{
- if (!page->age) {
- page->age = PAGE_AGE_START;
- ClearPageReferenced(page);
+ if (!PageActive(page)) {
+ if (page->age)
+ activate_page(page);
+ else {
+ page->age = PAGE_AGE_START;
+ ClearPageReferenced(page);
+ }
}
}
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: 2.4.7-ac4 disk thrashing
2001-08-08 6:38 2.4.7-ac4 disk thrashing Dieter Nützel
@ 2001-08-08 10:57 ` Alan Cox
2001-08-08 15:41 ` Daniel Phillips
0 siblings, 1 reply; 9+ messages in thread
From: Alan Cox @ 2001-08-08 10:57 UTC (permalink / raw)
To: Dieter Nützel
Cc: Linux Kernel List, ReiserFS List, Chris Mason, Nikita Danilov,
Daniel Phillips, Tom Vier
> Could it be that the ReiserFS cleanups in ac4 do harm?
> http://marc.theaimsgroup.com/?l=3Dreiserfs&m=3D99683332027428&w=3D2
I suspect the use once patch is the more relevant one.
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: 2.4.7-ac4 disk thrashing
@ 2001-08-08 6:38 Dieter Nützel
2001-08-08 10:57 ` Alan Cox
0 siblings, 1 reply; 9+ messages in thread
From: Dieter Nützel @ 2001-08-08 6:38 UTC (permalink / raw)
To: Linux Kernel List, ReiserFS List
Cc: Chris Mason, Nikita Danilov, Daniel Phillips, Tom Vier
Tom Vier wrote:
>switching from 2.4.7-ac3 to -ac4, disk access seems to be much more
>synchronis. running a ./configure script causes all kinds of trashing, as
>does installing .debs. i'm using reiserfs on top of software raid 0 on an
>alpha.
I can second that on 2.4.7-ac4 - ac9 (all versions).
Dbench show a dramatic decrease in disk troughput (~10 MB/sec) for every case
I've tested. I have a ReiserFS only system and the test partition was /opt
(/dev/sda8) with ~2.7 GB stuff on it, so we have some aging, too. It is the
last and slowest partition on my fast IBM U160 18 GB, 10.000 RPMs disk.
I've used 2.4.7 + acX + transaction-tracking-2 (Chris) + use-once-pages
(Daniel).
Now, some numbers (it should be noted that the u+s times are mostly equal but
the whole time and the throughput is different.
first lines for 2.4.7-ac3, then 2.4.7-ac9
dbench 48
Throughput 27.3983 MB/sec (NB=34.2479 MB/sec 273.983 MBit/sec)
37.580u 115.730s 3:51.30 66.2% 0+0k 0+0io 1310pf+0w
Throughput 18.4711 MB/sec (NB=23.0889 MB/sec 184.711 MBit/sec)
37.710u 121.900s 5:43.05 46.5% 0+0k 0+0io 1311pf+0w
dbench 32
Throughput 34.7552 MB/sec (NB=43.444 MB/sec 347.552 MBit/sec)
24.620u 73.980s 2:02.55 80.4% 0+0k 0+0io 911pf+0w
Throughput 21.8827 MB/sec (NB=27.3533 MB/sec 218.827 MBit/sec)
25.410u 76.610s 3:14.04 52.5% 0+0k 0+0io 912pf+0w
dbench 16
16 clients started
Throughput 37.7379 MB/sec (NB=47.1724 MB/sec 377.379 MBit/sec)
12.350u 35.330s 0:56.97 83.6% 0+0k 0+0io 511pf+0w
Throughput 30.0396 MB/sec (NB=37.5495 MB/sec 300.396 MBit/sec)
12.970u 37.320s 1:10.31 71.5% 0+0k 0+0io 511pf+0w
dbench 8
Throughput 40.9394 MB/sec (NB=51.1742 MB/sec 409.394 MBit/sec)
6.080u 17.420s 0:26.80 87.6% 0+0k 0+0io 311pf+0w
Throughput 28.174 MB/sec (NB=35.2175 MB/sec 281.74 MBit/sec)
6.280u 18.360s 0:38.49 64.0% 0+0k 0+0io 312pf+0w
dbench 4
Throughput 41.4035 MB/sec (NB=51.7544 MB/sec 414.035 MBit/sec)
3.140u 8.240s 0:13.76 82.7% 0+0k 0+0io 211pf+0w
Throughput 25.2641 MB/sec (NB=31.5801 MB/sec 252.641 MBit/sec)
3.270u 8.680s 0:21.91 54.5% 0+0k 0+0io 212pf+0w
dbench 2
Throughput 38.6387 MB/sec (NB=48.2983 MB/sec 386.387 MBit/sec)
1.680u 4.030s 0:07.83 72.9% 0+0k 0+0io 161pf+0w
Throughput 30.4352 MB/sec (NB=38.0441 MB/sec 304.352 MBit/sec)
1.690u 4.300s 0:09.68 61.8% 0+0k 0+0io 162pf+0w
dbench 1
Throughput 33.3689 MB/sec (NB=41.7111 MB/sec 333.689 MBit/sec)
0.820u 2.000s 0:04.96 56.8% 0+0k 0+0io 136pf+0w
Throughput 30.8583 MB/sec (NB=38.5729 MB/sec 308.583 MBit/sec)
0.750u 2.010s 0:05.28 52.2% 0+0k 0+0io 137pf+0w
System spec:
Athlon 550 I (yes, the first generation)
MSI MS-6167 Rev 1.0B (AMD Irongate C4, without bypass)
640 MB Pc100-2-2-2 SDRAM
SCSI subsystem driver Revision: 1.00
scsi0 : Adaptec AIC7XXX EISA/VLB/PCI SCSI HBA DRIVER, Rev 6.2.1
<Adaptec 2940 Ultra SCSI adapter>
aic7880: Ultra Wide Channel A, SCSI Id=7, 16/255 SCBs
Vendor: IBM Model: DDYS-T18350N Rev: S96H
Type: Direct-Access ANSI SCSI revision: 03
Vendor: IBM Model: DDRS-34560D Rev: DC1B
Type: Direct-Access ANSI SCSI revision: 02
Vendor: IBM Model: DDRS-34560W Rev: S71D
Type: Direct-Access ANSI SCSI revision: 02
Filesystem 1k-blocks Used Available Use% Mounted on
/dev/sda3 104412 69616 34796 67% /
/dev/sda2 1518088 37392 1480696 3% /tmp
/dev/sda5 1028092 334056 694036 33% /var
/dev/sda6 2048188 87188 1961000 5% /home
/dev/sda7 5124536 1752784 3371752 35% /usr
/dev/sda8 7068348 2715860 4352488 39% /opt
tmpfs 321188 0 321188 0% /dev/shm
Could it be that the ReiserFS cleanups in ac4 do harm?
http://marc.theaimsgroup.com/?l=reiserfs&m=99683332027428&w=2
Thanks,
Dieter
--
Dieter Nützel
Graduate Student, Computer Science
University of Hamburg
Department of Computer Science
Cognitive Systems Group
Vogt-Kölln-Straße 30
D-22527 Hamburg, Germany
email: nuetzel@kogs.informatik.uni-hamburg.de
@home: Dieter.Nuetzel@hamburg.de
^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2001-08-08 16:13 UTC | newest]
Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2001-08-04 15:38 2.4.7-ac4 disk thrashing Tom Vier
2001-08-04 23:04 ` Matthew Gardiner
2001-08-05 0:55 ` Tom Vier
2001-08-05 1:09 ` ip_conntrack bigger, better and yellow ;) !!! Mircea Ciocan
2001-08-05 3:19 ` Sven Koch
2001-08-08 6:38 2.4.7-ac4 disk thrashing Dieter Nützel
2001-08-08 10:57 ` Alan Cox
2001-08-08 15:41 ` Daniel Phillips
2001-08-08 16:03 ` Dieter Nützel
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).