Re: 2.4.10aa1 (00_vm-tweaks-1)

Re: 2.4.10aa1 (00_vm-tweaks-1)

[arjan]

In article <20010923232511.B1466@athlon.random> you wrote:

> Only in 2.4.10aa1: 00_enable-apic-1

>        Enable UP ioapic on demand via boot param.

If you took my patch for it, PLEASE don't send it for inclusion; it's an
evil hack and no longer needed when Intel fixes the bug in their 440GX bios.

Greetings,
   Arjan van de Ven

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Alan Cox]

> If you took my patch for it, PLEASE don't send it for inclusion; it's an
> evil hack and no longer needed when Intel fixes the bug in their 440GX bios.

"when" is not a word I find useful about most bios bugs. Try "if" or
"less likely that being hit on the head by an asteroid"

Alan

Re: 2.4.10aa1 (00_vm-tweaks-1)

[arjan]

In article <E15lH4e-0000VP-00@the-village.bc.nu> you wrote:
>> If you took my patch for it, PLEASE don't send it for inclusion; it's an
>> evil hack and no longer needed when Intel fixes the bug in their 440GX bios.

> "when" is not a word I find useful about most bios bugs. Try "if" or
> "less likely that being hit on the head by an asteroid"

This case is different. Intel promised the update "soon". For them it's
simple, either they update or they can't say their 440GX boards are linux
compatible, because with the current bios they really aren't. 

They could also chose to open the "Intel Proprietary" information that is
needed to actually work with this board properly instead of via bios hacks,
but that's their choice...

Greetings,
   Arjan van de Ven

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Andrea Arcangeli]

On Sun, Sep 23, 2001 at 10:36:08PM +0100, arjan@fenrus.demon.nl wrote:
> In article <20010923232511.B1466@athlon.random> you wrote:
> 
> > Only in 2.4.10aa1: 00_enable-apic-1
> 
> >        Enable UP ioapic on demand via boot param.
> 
> If you took my patch for it, PLEASE don't send it for inclusion; it's an
> evil hack and no longer needed when Intel fixes the bug in their 440GX bios.

we're having troubles with the ioapic compiled in in UP on some machine,
on 2.4.10pre I assume. Waiting bios updates is more problematic then
enabling on demand rather than disabling on demand, so yes I will be
obviously glad to drop this one as soon as the problem goes away.
thanks for the info.

Andrea

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Andrea Arcangeli]

On Sun, Sep 23, 2001 at 10:46:03PM +0100, Alan Cox wrote:
> "less likely that being hit on the head by an asteroid"

:)

Andrea

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Adrian Bunk]

On Sun, 23 Sep 2001, Andrea Arcangeli wrote:

> If you are interested about the VM behaviour under swap (and non) please
> test the 00_vm-tweaks-1 (can be applied to plain 2.4.10), here the swap
> behaviour seems improved with it. Thanks!
>...

2.4.10 seems to behave worse than kernels up to 2.4.9 and 2.4.9ac12
(haven't tried above) with the following workload (everything is only a
subjective impression as a user; I don't look at how long the "rm" and the
"tar" take because that's not very important for me):

FVWM with 6 open xterms is running
XMMS is running

mv linux linux.old
nice rm -rf linux.old &
tar xzf linux-2.4.10.tar.gz &
lynx ftp://ftp.kernel.org/pub/linux/kernel/testing


up to 2.4.9 and 2.4.9ac12:
everything works fine

2.4.10:
interactive use of the machine is very bad, I can't type a command in
another xterm

2.4.10 + 00_vm-tweaks-1:
interactive use of the machine works all right, but XMMS does sometimes
stutter


This is on a K6 with 300 Mhz. free in 2.4.10 + 00_vm-tweaks-1 gives the
following output:

$ free
             total       used       free     shared    buffers     cached
Mem:         62252      60400       1852          0       2408      32792
-/+ buffers/cache:      25200      37052
Swap:       947824      10264     937560
$


My impression is that there's a problem when several processes do heavy
non-cachable disk IO.


cu
Adrian

-- 

Get my GPG key: finger bunk@debian.org | gpg --import

Fingerprint: B29C E71E FE19 6755 5C8A  84D4 99FC EA98 4F12 B400

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Ingo Molnar]

On Sun, 23 Sep 2001, Andrea Arcangeli wrote:

> we're having troubles with the ioapic compiled in in UP on some
> machine, on 2.4.10pre I assume. Waiting bios updates is more
> problematic then enabling on demand rather than disabling on demand,

the right policy is to disable-on-demand, especially if the breakage is
well established and is also uncommon. It's the faulty case that should do
the extra work. Good systems should not be punished needlessly.

	Ingo

Re: 2.4.10aa1 (00_vm-tweaks-1)

[john slee]

On Sun, Sep 23, 2001 at 10:46:36PM +0100, arjan@fenrus.demon.nl wrote:
> In article <E15lH4e-0000VP-00@the-village.bc.nu> you wrote:
> >> If you took my patch for it, PLEASE don't send it for inclusion; it's an
> >> evil hack and no longer needed when Intel fixes the bug in their 440GX bios.
> 
> > "when" is not a word I find useful about most bios bugs. Try "if" or
> > "less likely that being hit on the head by an asteroid"
> 
> This case is different. Intel promised the update "soon". For them it's

oh, they _promised_!  great!

on another note...

2.4.10 is great incidentally, noticeably improved performance for
a somewhat large postgresql db on raid0 (35GB, ~100 million rows)

if anyone's interested i'll post numbers vs. 2.4.9 as i do more testing.

/me bearhug andrea

j.

-- 
R N G G   "Well, there it goes again... And we just sit 
 I G G G   here without opposable thumbs." -- gary larson

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Marcelo Tosatti]

On Tue, 25 Sep 2001, john slee wrote:

> On Sun, Sep 23, 2001 at 10:46:36PM +0100, arjan@fenrus.demon.nl wrote:
> > In article <E15lH4e-0000VP-00@the-village.bc.nu> you wrote:
> > >> If you took my patch for it, PLEASE don't send it for inclusion; it's an
> > >> evil hack and no longer needed when Intel fixes the bug in their 440GX bios.
> > 
> > > "when" is not a word I find useful about most bios bugs. Try "if" or
> > > "less likely that being hit on the head by an asteroid"
> > 
> > This case is different. Intel promised the update "soon". For them it's
> 
> oh, they _promised_!  great!
> 
> on another note...
> 
> 2.4.10 is great incidentally, noticeably improved performance for
> a somewhat large postgresql db on raid0 (35GB, ~100 million rows)
> 
> if anyone's interested i'll post numbers vs. 2.4.9 as i do more testing.

Please do. (the more detailed the better)

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Andrea Arcangeli]

On Tue, Sep 25, 2001 at 06:33:57PM +1000, john slee wrote:
> On Sun, Sep 23, 2001 at 10:46:36PM +0100, arjan@fenrus.demon.nl wrote:
> > In article <E15lH4e-0000VP-00@the-village.bc.nu> you wrote:
> > >> If you took my patch for it, PLEASE don't send it for inclusion; it's an
> > >> evil hack and no longer needed when Intel fixes the bug in their 440GX bios.
> > 
> > > "when" is not a word I find useful about most bios bugs. Try "if" or
> > > "less likely that being hit on the head by an asteroid"
> > 
> > This case is different. Intel promised the update "soon". For them it's
> 
> oh, they _promised_!  great!
> 
> on another note...
> 
> 2.4.10 is great incidentally, noticeably improved performance for
> a somewhat large postgresql db on raid0 (35GB, ~100 million rows)

This was actually the first priority and it's nice to hear it is
apparently been achieved successfully.

I also take the opportunity to remind everybody to keep the
00_vm-tweaks-1 from 2.4.10aa1 applied on top of 2.4.10, it seems quite
important here.

thanks,
Andrea

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Andrea Arcangeli]

On Mon, Sep 24, 2001 at 03:42:59PM +0200, Adrian Bunk wrote:
> On Sun, 23 Sep 2001, Andrea Arcangeli wrote:
> 
> > If you are interested about the VM behaviour under swap (and non) please
> > test the 00_vm-tweaks-1 (can be applied to plain 2.4.10), here the swap
> > behaviour seems improved with it. Thanks!
> >...
> 
> 2.4.10 seems to behave worse than kernels up to 2.4.9 and 2.4.9ac12
> (haven't tried above) with the following workload (everything is only a
> subjective impression as a user; I don't look at how long the "rm" and the
> "tar" take because that's not very important for me):
> 
> FVWM with 6 open xterms is running
> XMMS is running
> 
> mv linux linux.old
> nice rm -rf linux.old &
> tar xzf linux-2.4.10.tar.gz &
> lynx ftp://ftp.kernel.org/pub/linux/kernel/testing
> 
> 
> up to 2.4.9 and 2.4.9ac12:
> everything works fine
> 
> 2.4.10:
> interactive use of the machine is very bad, I can't type a command in
> another xterm

yes, that's because you didn't applied vm-tweaks-1 yet and you didn't
enough ram and you needed to swap.

> 
> 2.4.10 + 00_vm-tweaks-1:
> interactive use of the machine works all right, but XMMS does sometimes
> stutter

it didn't happen here but let's look into it.

> This is on a K6 with 300 Mhz. free in 2.4.10 + 00_vm-tweaks-1 gives the
> following output:
> 
> $ free
>              total       used       free     shared    buffers     cached
> Mem:         62252      60400       1852          0       2408      32792
> -/+ buffers/cache:      25200      37052
> Swap:       947824      10264     937560
> $
> 
> 
> My impression is that there's a problem when several processes do heavy
> non-cachable disk IO.

I've a few ideas on what to change incrementally to vm-tweaks-1, but can
you send me the `vmstat 1` output, plus also the /proc/meminfo output
snapshotted at regular intervals while xmms is "stuttering" just to
avoid looking at the wrong place? thanks,

Andrea

Re: 2.4.10aa1 (00_vm-tweaks-1)

[Adrian Bunk]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1220 bytes --]

On Tue, 25 Sep 2001, Andrea Arcangeli wrote:

>...
> > FVWM with 6 open xterms is running
> > XMMS is running
> >
> > mv linux linux.old
> > nice rm -rf linux.old &
> > tar xzf linux-2.4.10.tar.gz &
> > lynx ftp://ftp.kernel.org/pub/linux/kernel/testing
>...
> > 2.4.10 + 00_vm-tweaks-1:
> > interactive use of the machine works all right, but XMMS does sometimes
> > stutter
>
> it didn't happen here but let's look into it.
>
> > This is on a K6 with 300 Mhz. free in 2.4.10 + 00_vm-tweaks-1 gives the
>...
> > My impression is that there's a problem when several processes do heavy
> > non-cachable disk IO.
>
> I've a few ideas on what to change incrementally to vm-tweaks-1, but can
> you send me the `vmstat 1` output, plus also the /proc/meminfo output
> snapshotted at regular intervals while xmms is "stuttering" just to
> avoid looking at the wrong place? thanks,

Attached are:

vmstat: `vmstat 1` output continuously from the beginning of the test
(includes several stuttering periods)

meminfo: /proc/meminfo output of three times when I heard it stuttering
(XMMS doesn't stutter continuously but it has stuttering periods that last
a few seconds - I hope I catched the right moments)


> Andrea

cu
Adrian

[-- Attachment #2: Type: TEXT/PLAIN, Size: 1608 bytes --]

        total:    used:    free:  shared: buffers:  cached:
Mem:  63746048 61468672  2277376        0  3018752 25452544
Swap: 970571776 12042240 958529536
MemTotal:        62252 kB
MemFree:          2224 kB
MemShared:           0 kB
Buffers:          2948 kB
Cached:          24764 kB
SwapCached:         92 kB
Active:          14392 kB
Inactive:        13412 kB
HighTotal:           0 kB
HighFree:            0 kB
LowTotal:        62252 kB
LowFree:          2224 kB
SwapTotal:      947824 kB
SwapFree:       936064 kB
        total:    used:    free:  shared: buffers:  cached:
Mem:  63746048 61231104  2514944        0  2768896 25587712
Swap: 970571776 11952128 958619648
MemTotal:        62252 kB
MemFree:          2456 kB
MemShared:           0 kB
Buffers:          2704 kB
Cached:          24984 kB
SwapCached:          4 kB
Active:          12652 kB
Inactive:        15040 kB
HighTotal:           0 kB
HighFree:            0 kB
LowTotal:        62252 kB
LowFree:          2456 kB
SwapTotal:      947824 kB
SwapFree:       936152 kB
        total:    used:    free:  shared: buffers:  cached:
Mem:  63746048 61710336  2035712        0  2719744 25526272
Swap: 970571776 11952128 958619648
MemTotal:        62252 kB
MemFree:          1988 kB
MemShared:           0 kB
Buffers:          2656 kB
Cached:          24924 kB
SwapCached:          4 kB
Active:          12664 kB
Inactive:        14920 kB
HighTotal:           0 kB
HighFree:            0 kB
LowTotal:        62252 kB
LowFree:          1988 kB
SwapTotal:      947824 kB
SwapFree:       936152 kB

[-- Attachment #3: Type: TEXT/PLAIN, Size: 5913 bytes --]

   procs                      memory    swap          io     system         cpu
 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  id
 2  1  1  11532   2164   3484  24908   0   0    10    10  109    78   1   0  98
 3  1  0  11532   2312   3440  24692   0   0  1167     0  256  2092  44  29  26
 2  2  0  11532   2100   3432  24524   0   0   570     0  214  2717  55  41   4
 2  3  0  11532   2296   3252  24500   0   0   984   271  249  2404  54  40   6
 1  4  1  11532   2220   3228  24596   0   0    47   518  279  1292  14  11  74
 1  4  1  11532   2432   3156  24436   0   0   298   897  230  1973  31  20  49
 2  3  2  11532   2020   3152  24848   0   0   595  1991  220  2014  50  29  20
 1  4  2  11760   1208   3120  25792   0   0   531  3828  256  1932  46  35  19
 3  2  1  11760   1728   3104  25264   0   0   131  1046  287  1059  15  12  73
 3  2  1  11760   1916   3024  25096   0 140   466  1510  248  1427  35  20  45
 3  4  2  11760   2232   2980  24784   0   0   502  1914  271  1261  28  19  53
 1  3  2  11760   1816   2948  25336   0   0   408  1675  283  1177  28  16  56
 4  2  2  11760   2244   2904  24940   0   0   644   918  254  1557  41  33  26
 0  4  1  11760   2312   2888  24884   0   0   487  1548  241  2010  66  25   9
 4  1  2  11760   2304   2904  24872   0   0   561   998  226  1559  27  22  51
 3  2  1  11672   2312   2816  25272   0   0   743   893  226  2181  54  33  12
 2  3  3  11672   2096   2800  25456   0   0   364  1546  270  1544  34  16  50
 3  2  3  11672   1824   2764  25736   0   0   270   847  272  1203  18  14  67
 3  3  2  11672   2464   2704  24984   0   0   400   682  260  1177  22  17  61
 2  2  1  11672   1980   2732  25436   0   0   743   914  257  1858  44  29  27
 2  3  1  11672   1908   2760  25372   0   0   476  2311  259  1407  23  27  50
   procs                      memory    swap          io     system         cpu
 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  id
 3  1  2  11672   1852   2764  25160   0   0   442  1043  258  1349  26  20  54
 1  3  1  11672   1800   2780  25168   0   0   594  1232  281  1482  36  19  45
 0  4  2  11672   1884   2764  25100   0   0    37   381  304   783   9   8  84
 0  4  1  11672   1856   2776  25116   0   0   253   418  227   799  13   6  82
 0  4  1  11672   2276   2748  24708   0   0   202  1251  333   927   9  12  80
 0  4  1  11672   2236   2696  24800   0   0   221  1005  230   704   6   6  88
 1  3  2  11672   2432   2664  24596   0   0   464  1343  251  1713  38  21  41
 2  4  2  11672   1976   2656  24940   0   0   548   524  320  1368  74  21   5
 2  3  1  11672   2252   2664  24732   0   0   170   645  288  1170  34   8  58
 3  2  1  11672   1848   2680  25100   0   0   314   160  285  1432  20  11  69
 4  1  2  11672   1916   2672  25024   0   0   166   946  246  1406  15  16  68
 1  3  1  11672   2212   2324  25040   0   0   491  2865  250  2142  44  36  20
 1  3  2  11672   1976   2028  25536   0   0   518  1113  327  2071  37  31  33
 1  3  1  11672   2048   1952  25528   0   0   299  1094  307  1583  24  22  54
 2  3  1  11672   1888   1976  25572   0   0   432  1152  268  1581  34  19  47
 0  4  1  11672   1872   2020  25488   0   0   624  2113  244  1871  39  24  37
 2  2  1  11672   2312   1912  25108   0   0   596  1536  238  1415  36  18  46
 0  3  1  11672   1892   1900  25372   0   0   543  1860  308  1857  56  28  16
 0  3  1  11672   1936   1924  25272   0   0   419  1952  204  1673  48  20  32
 1  2  1  11672   2184   1928  25004   0   0   417  1570  252  1486  39  20  41
 3  0  2  11672   2144   1972  24944   0   0   548  2112  258  2102  49  28  23
   procs                      memory    swap          io     system         cpu
 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  id
 1  2  1  11672   2528   1964  24564   0   0   660  1573  246  1756  79  17   4
 2  2  2  11672   2300   1928  24784   0   0   683  2369  258  2253  61  35   4
 1  2  1  11672   1944   1948  25088   0   0   435  2080  216  2068  44  32  24
 2  2  1  11672   2576   1964  24412   0   0   460  1765  208  1864  47  23  30
 5  0  2  11672   2168   1968  24772   0   0   523  1792  221  1887  37  27  37
 2  2  2  11672   2156   1932  24792   0   0   421  1762  218  1874  45  23  32
 0  3  1  11672   2696   1820  24328   0   0   565  1602  269  1853  46  20  34
 3  0  2  11672   2112   1828  24872   0   0   413  1220  219  1656  75  19   7
 2  2  1  11672   2092   1868  24760   0   0   411  2113  248  2071  45  37  18
 2  2  2  11672   1812   1808  25044   0   0   556  2019  287  2437  50  32  18
 2  2  1  11672   2248   1824  24552   0   0   322  1408  224  2350  45  22  33
 2  2  1  11672   2272   1828  24512   0   0   307  1408  252  2465  55  22  23
 3  1  1  11672   2328   1824  24424   0   0   317  1284  242  2084  50  19  31
 1  3  1  11672   3344   1832  24308   0   0   422  1056  233  1628  40  17  43
 0  3  1  11672   2008   1880  25536   0   0   561  1667  237  2014  47  25  28
 2  2  2  11672   2256   1868  25272   0   0   288  1568  247  1790  29  27  45
 4  0  1  11672   2132   1912  25312   0   0   445  1344  222  1726  53  15  31
 0  3  1  11672   2144   1948  25232   0   0   441  1477  252  1777  38  28  35
 0  3  1  11672   1992   1964  25304   0   0   283  1504  281  1723  31  21  48
 5  0  2  11672   1868   1984  25368   0   0   530  1345  239  2003  67  19  14
 3  1  2  11672   2284   1968  24900   0   0   554  1792  299  1928  77  22   1
   procs                      memory    swap          io     system         cpu
 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy  id
 2  2  1  11672   2408   1996  24648   0   0   281  1797  224  1972  31  30  39
 2  2  2  11672   2160   2040  24780   0   0   436  1728  271  2382  50  26  24