Skip to contentSkip to conversation

Read this and be ashamed ;) or: Awfull performance loss since 2.4.18 to 2.4.21-pre2

DISCUSSION Marc-Christian Petersen 4 participants 8 messages Dec 22, 2002 – Dec 23, 2002

lore
Share & tools
lore
#0 lore raw

Hi all,

not much to say about, just read. All are vanilla kernels w/o any patch.

/dev/hda5 on /home type ext3 (rw,data=ordered)
/dev/hda5 10080488 731488 8836932 8% /home

UDMA100 IDE Drive, DMA is on. All these runs were done right after bootup. Mashine is a Celeron 1,3GHz, 512MB RAM, 512MB SWAP.

root@codeman:[/] # uname -r
2.4.18
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 119.140681 seconds (18024772 bytes/sec) root@codeman:[/] #

root@codeman:[/] # uname -r
2.4.19
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 140.305836 seconds (15305733 bytes/sec)

root@codeman:[/] # uname -r
2.4.20
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 172.327570 seconds (12461637 bytes/sec)

root@codeman:[/] # uname -r
2.4.21-pre2
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 177.743959 seconds (12081894 bytes/sec)

ciao, Marc

#1 lore raw

On Sunday 22 December 2002 14:47, Marc-Christian Petersen wrote:

Hi again,

Quoted 6 lines

root@codeman:[/] # uname -r
2.4.20
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 172.327570 seconds (12461637 bytes/sec)

root@codeman:[/] # uname -r
2.4.20-rmap15b
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 140.460427 seconds (15288887 bytes/sec)

ciao, Marc

#2 lore raw

On Sunday 22 December 2002 15:38, Marc-Christian Petersen wrote:

And hi again ^3,

Quoted 6 lines

root@codeman:[/] # uname -r
2.4.20-rmap15b
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 140.460427 seconds (15288887 bytes/sec)

root@codeman:[/] # uname -r
2.4.20aa1
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 286.054011 seconds (7507266 bytes/sec)

ciao, Marc

#3 lore raw
Quoted 11 lines · On 2002.12.22 Marc-Christian Petersen wrote:

On Sunday 22 December 2002 15:38, Marc-Christian Petersen wrote:

And hi again ^3,

Quoted 6 lines

root@codeman:[/] # uname -r
2.4.20-rmap15b
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 140.460427 seconds (15288887 bytes/sec)

Quoted 7 lines

root@codeman:[/] # uname -r
2.4.20aa1
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 286.054011 seconds (7507266 bytes/sec)

Check you timer...

werewolf:~> uname -a
Linux werewolf.able.es 2.4.20-jam2 #2 SMP vie dic 20 22:35:33 CET 2002 i686 unknown unknown GNU/Linux werewolf:~> time dd if=/dev/zero of=kk bs=16384 count=131072
131072+0 records in
131072+0 records out
0.51user 60.63system 1:23.73elapsed 73%CPU (0avgtext+0avgdata 0maxresident)k 0inputs+0outputs (133major+15minor)pagefaults 0swaps

Box is a dual PII@400, 950Mb of RAM. FS is ext3.
So about 83 seconds on -jam2, which is mainly just 2.4.20aa1 with ext3 fixes. Ah, no special options to ext3 mount (no data=ordered). That's the point ?

???

Signature · 4 lines
J.A. Magallon <[email protected]>      \                 Software is like sex:
werewolf.able.es                         \           It's better when it's free
Mandrake Linux release 9.1 (Cooker) for i586
Linux 2.4.20-jam2 (gcc 3.2.1 (Mandrake Linux 9.1 3.2.1-2mdk))
#4 lore raw

On Sunday 22 December 2002 23:21, J.A. Magallon wrote:

Hi J.A.

Quoted 12 lines · On 2002.12.22 Marc-Christian Petersen wrote:
Quoted 11 lines · Marc-Christian Petersen

And hi again ^3,

Quoted 6 lines

root@codeman:[/] # uname -r
2.4.20-rmap15b
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 140.460427 seconds (15288887 bytes/sec)

Quoted 8 lines
Quoted 6 lines

root@codeman:[/] # uname -r
2.4.20aa1
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 286.054011 seconds (7507266 bytes/sec)

Check you timer...

the timer is right.

Quoted 4 lines

Box is a dual PII@400, 950Mb of RAM. FS is ext3.
So about 83 seconds on -jam2, which is mainly just 2.4.20aa1 with ext3 fixes. Ah, no special options to ext3 mount (no data=ordered). That's the point ?

nono, ordered is default if nothing is specified.

Seems it makes a big difference with DUAL and 1GB RAM. Now with Dual P3 1GHz, 1GB RAM, ext3 ordered mode:

root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 55.741130 seconds (38526016 bytes/sec)

Anyway, the comparison to the different kernels shows up again a better throughput than with -aa. That is pretty obvious because of the lowlatency elevator in -aa. But that's not the point. The point is that with every new kernel release the performance drops. So it shows by dd. Not to mention the pauses/stops grow up with every new release. It cannot be true that compared 2.4.18 to 2.4.20 there is a 6mb/s performance loss. What the hell we can get with $MONSTER-BOX w/o the performance issues ;) Anyway, that dd call with those options is pretty expensive to mem_load.

I wonder why this doesn't affect contest [tm]. Or am I blind?

ciao, Marc

#5 lore raw
In reply to Marc-Christian Petersen
Quoted 5 lines

2.4.18
2147483648 bytes transferred in 119.140681 seconds (18024772 bytes/sec)

2.4.19
2147483648 bytes transferred in 140.305836 seconds (15305733 bytes/sec)

Well I'm getting the numbers the other way round. <g>

Machine:

$ egrep 'model name|MHz' /proc/cpuinfo
model name      : AMD Athlon(tm) processor
cpu MHz         : 996.037
$ grep MemTotal /proc/meminfo
MemTotal:       516588 kB
$ egrep 'Bridge|DMA rate' /proc/ide/via
South Bridge:                       VIA vt82c686b
Highest DMA rate:                   UDMA100
$ cat /proc/ide/ide0/hda/model
ST380021A
$ su -c 'hdparm -Iv /dev/hda| grep -i dma'
 using_dma    =  1 (on)
        DMA: mdma0 mdma1 mdma2 udma0 udma1 udma2 udma3 udma4 *udma5
$ grep [[:blank:]]/[[:blank:]] /etc/fstab
/dev/hda3	/	reiserfs	defaults	1	1

Tests:

$ uname -r
2.4.18
$ time sh -c 'sync; sync; sync; dd if=/dev/zero of=hoax bs=16k count=131072; sync; sync; sync'
131072+0 records in
131072+0 records out
real    1m44.708s
user    0m0.140s
sys     0m20.340s
=> 19.56MB/s
$ uname -r
2.4.20
$ time sh -c 'sync; sync; sync; dd if=/dev/zero of=hoax bs=16k count=131072; sync; sync; sync'
131072+0 records in
131072+0 records out
real    1m27.980s
user    0m0.120s
sys     0m20.290s
=> 23.28MB/s

I also tried machines with disks connected to various SCSI controllers and in all cases more recent kernels gave better results than older ones (sym53c8xx: two-disk raid1 - 2.4.18: ~35MB/s, 2.4.20: ~40MB/s). I'm using reiserfs 3.6 everywhere.

Signature · 1 line
Tomas Szepe <[email protected]>
#6 lore raw

On Monday 23 December 2002 13:03, Tomas Szepe wrote:

Hi Tomas,

Quoted 7 lines
Quoted 5 lines

2.4.18
2147483648 bytes transferred in 119.140681 seconds (18024772 bytes/sec)

2.4.19
2147483648 bytes transferred in 140.305836 seconds (15305733 bytes/sec)

Well I'm getting the numbers the other way round. <g>

lol, strange.

root@codeman:[/] # egrep 'model name|MHz' /proc/cpuinfo
model name : Intel(R) Celeron(TM) CPU 1300MHz
cpu MHz : 1295.718

root@codeman:[/] # grep MemTotal /proc/meminfo
MemTotal: 515056 kB

root@codeman:[/proc/ide] # egrep 'Intel|UDMA enabled' /proc/ide/piix
Intel PIIX4 Ultra 100 Chipset.

UDMA enabled:   yes              no              no                no 
UDMA enabled:   5                X               X                 X

root@codeman:[/] # hdparm -i /dev/hda

/dev/hda:

Model=MAXTOR 6L060J3, FwRev=A93.0500, SerialNo=663219752652 Config={ HardSect NotMFM HdSw>15uSec Fixed DTR>10Mbs } RawCHS=16383/16/63, TrkSize=32256, SectSize=21298, ECCbytes=4 BuffType=DualPortCache, BuffSize=1819kB, MaxMultSect=16, MultSect=16 CurCHS=4047/16/255, CurSects=16511760, LBA=yes, LBAsects=117266688 IORDY=on/off, tPIO={min:120,w/IORDY:120}, tDMA={min:120,rec:120} PIO modes: pio0 pio1 pio2 pio3 pio4
DMA modes: mdma0 mdma1 mdma2
UDMA modes: udma0 udma1 udma2 udma3 udma4 *udma5 udma6 AdvancedPM=no WriteCache=enabled
Drive conforms to: ATA/ATAPI-5 T13 1321D revision 1: 1 2 3 4 5

ciao, Marc

#7 lore raw
In reply to Marc-Christian Petersen

hi

What disk/controller is this?

Quoted 46 lines · Marc-Christian Petersen

Hi all,

not much to say about, just read. All are vanilla kernels w/o any patch.

/dev/hda5 on /home type ext3 (rw,data=ordered)
/dev/hda5 10080488 731488 8836932 8% /home

UDMA100 IDE Drive, DMA is on. All these runs were done right after bootup. Mashine is a Celeron 1,3GHz, 512MB RAM, 512MB SWAP.

root@codeman:[/] # uname -r
2.4.18
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 119.140681 seconds (18024772 bytes/sec) root@codeman:[/] #

root@codeman:[/] # uname -r
2.4.19
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 140.305836 seconds (15305733 bytes/sec)

root@codeman:[/] # uname -r
2.4.20
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 172.327570 seconds (12461637 bytes/sec)

root@codeman:[/] # uname -r
2.4.21-pre2
root@codeman:[/] # dd if=/dev/zero of=/home/largefile bs=16384 count=131072 131072+0 records in
131072+0 records out
2147483648 bytes transferred in 177.743959 seconds (12081894 bytes/sec)

ciao, Marc
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/