Question on IICACHE and mount info (2.4.19-pre5)

Question on IICACHE and mount info (2.4.19-pre5)

[Manuel Krause]

Hi!

Is there only the addidional mount option "iicache" with this patch:
ftp://ftp.namesys.com/pub/reiserfs-for-2.4/testing/linux-2.4.19p4-iicache-new-5.patch 
?

There were some more in the early versions of iicache, IIRC.

But at the moment it works very well with 2.4.19-pre5 + 
rml.preempt+lock-break + aa.vm-33 + some other aa patches and some akpm 
ones.

When I put ",iicache" to the fstab lines of my reiserfs partitions I 
don't get a response from

# cat "/proc/fs/reiserfs/ide0(3,11)/super"
state:  REISERFS_VALID_FS
mount options:  BORDER NO_TAILS LOG
...

# cat "/proc/fs/reiserfs/ide0(3,7)/super"
state:  REISERFS_VALID_FS
mount options:  BORDER NO_TAILS LOG
...

# cat /proc/mounts
/dev/root / reiserfs rw,noatime 0 0
/dev/hda11 /mnt/beta reiserfs rw,noatime,nosuid,nodev 0 0
...

only # mount says:
/dev/hda7 on / type reiserfs (rw,noatime,notail,iicache)
...
/dev/hda11 on /mnt/beta type reiserfs 
(rw,nosuid,nodev,noatime,notail,iicache)
...

Seems like some things have gone with the latest patches which worked 
before? It's o.k., of course, if "iicache" doesn't show up for now, I 
know it's for testing only (yes, it would be very nice!). But the 
"notail" thing worked before!


Best regards,

Manuel

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Oleg Drokin]

Hello!

On Fri, Apr 05, 2002 at 02:00:03AM +0200, Manuel Krause wrote:

> Is there only the addidional mount option "iicache" with this patch:
> ftp://ftp.namesys.com/pub/reiserfs-for-2.4/testing/linux-2.4.19p4-iicache-new-5.patch 
> ?
> There were some more in the early versions of iicache, IIRC.

most of other options were considered unsafe, so probably they were dropped.

> But at the moment it works very well with 2.4.19-pre5 + 
> rml.preempt+lock-break + aa.vm-33 + some other aa patches and some akpm 
> ones.

Do you see any difference with and without iicache?

> When I put ",iicache" to the fstab lines of my reiserfs partitions I 
> don't get a response from
> # cat "/proc/fs/reiserfs/ide0(3,11)/super"
> state:  REISERFS_VALID_FS
> mount options:  BORDER NO_TAILS LOG
> # cat /proc/mounts
> /dev/root / reiserfs rw,noatime 0 0
> /dev/hda11 /mnt/beta reiserfs rw,noatime,nosuid,nodev 0 0
> only # mount says:
> /dev/hda7 on / type reiserfs (rw,noatime,notail,iicache)
> ...
> /dev/hda11 on /mnt/beta type reiserfs 
> (rw,nosuid,nodev,noatime,notail,iicache)
> ...

It seems this is lost piece of code, that prints this info.

> Seems like some things have gone with the latest patches which worked 
> before? It's o.k., of course, if "iicache" doesn't show up for now, I 
> know it's for testing only (yes, it would be very nice!). But the 
> "notail" thing worked before!

Which "notail" thing?

Bye,
    Oleg

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Manuel Krause]

On 04/05/2002 07:02 AM, Oleg Drokin wrote:

> Hello!
> 
> On Fri, Apr 05, 2002 at 02:00:03AM +0200, Manuel Krause wrote:
> 
> 
>>Is there only the addidional mount option "iicache" with this patch:
>>ftp://ftp.namesys.com/pub/reiserfs-for-2.4/testing/linux-2.4.19p4-iicache-new-5.patch 
>>?
>>There were some more in the early versions of iicache, IIRC.
>>
> 
> most of other options were considered unsafe, so probably they were dropped.
> 
> 
>>But at the moment it works very well with 2.4.19-pre5 + 
>>rml.preempt+lock-break + aa.vm-33 + some other aa patches and some akpm 
>>ones.
>>
> 
> Do you see any difference with and without iicache?
> 


At the moment I'm facing some stability problems with the other patches 
I applied. I could say more about differences after solving this. ;-)

> 
>>When I put ",iicache" to the fstab lines of my reiserfs partitions I 
>>don't get a response from
>># cat "/proc/fs/reiserfs/ide0(3,11)/super"
>>state:  REISERFS_VALID_FS
>>mount options:  BORDER NO_TAILS LOG
>># cat /proc/mounts
>>/dev/root / reiserfs rw,noatime 0 0
>>/dev/hda11 /mnt/beta reiserfs rw,noatime,nosuid,nodev 0 0
>>only # mount says:
>>/dev/hda7 on / type reiserfs (rw,noatime,notail,iicache)
>>...
>>/dev/hda11 on /mnt/beta type reiserfs 
>>(rw,nosuid,nodev,noatime,notail,iicache)
>>...
>>
> 
> It seems this is lost piece of code, that prints this info.
> 

> 
>>Seems like some things have gone with the latest patches which worked 
>>before? It's o.k., of course, if "iicache" doesn't show up for now, I 
>>know it's for testing only (yes, it would be very nice!). But the 
>>"notail" thing worked before!
>>
> 
> Which "notail" thing?
> 


Mmmh. I only repeated myself? I wanted to say that the "notail" mount 
option showed up in "# cat /proc/mounts" some weeks ago, IIRC.

> Bye,
>     Oleg
> 
> 


I'll try the latest one (linux-2.4.19p5-iicache-new-10.patch) instead of -5.

Thanks,

Manuel

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Oleg Drokin]

Hello!

On Fri, Apr 05, 2002 at 03:27:03PM +0200, Manuel Krause wrote:
> >>Seems like some things have gone with the latest patches which worked 
> >>before? It's o.k., of course, if "iicache" doesn't show up for now, I 
> >>know it's for testing only (yes, it would be very nice!). But the 
> >>"notail" thing worked before!
> >Which "notail" thing?
> Mmmh. I only repeated myself? I wanted to say that the "notail" mount 
> option showed up in "# cat /proc/mounts" some weeks ago, IIRC.

Actually function to implement correct /proc/mounts output
for reiserfs never was implemented, though I think I will return to it.
But right now there are more important stuff to do.

Bye,
    Oleg

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Manuel Krause]

On 04/05/2002 07:02 AM, Oleg Drokin wrote:

> Hello!
> 
> On Fri, Apr 05, 2002 at 02:00:03AM +0200, Manuel Krause wrote:
> 
[...]
> 
> Do you see any difference with and without iicache?
> 
[...]
> 
> Bye,
>     Oleg
> 


So, I made some semi-professional tests (only one run, and I have no 
overview about my files size statistics) with my reiserfs partitions. 
The data is only copied from the partitions backup counterpart that was 
freshly created. There was KDE up during copying.


First partition: 1958788 kB (my system)
---------------------------------------
notail:
 real    5m16.853s user    0m1.660s sys     0m53.090s

notail+noborder:
 real    5m30.247s user    0m1.580s sys     0m49.840s

notail+iicache:
 real    4m55.041s user    0m1.820s sys     0m52.650s

notail+noborder+iicache:
 real    5m32.181s user    0m1.260s sys     0m48.590s

 
Second partition: 4085252 kB (mostly archives)
------------------------------------------
notail:
 real    11m3.024s  user    0m1.460s sys     1m26.780s

notail+noborder:
 real    12m38.724s user    0m1.380s sys     1m22.570s

notail+iicache:
 real    11m13.734s user    0m1.620s sys     1m26.650s

notail+noborder+iicache:
 real    11m50.198s user    0m1.520s sys     1m19.530s



Yes, I have differences. But I think they're not the ones we wanted to 
see?! I double checked the lines I pasted to this email ;-)


Best regards,

Manuel

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Oleg Drokin]

Hello!

On Mon, Apr 08, 2002 at 01:44:48PM +0200, Manuel Krause wrote:

> So, I made some semi-professional tests (only one run, and I have no 
> overview about my files size statistics) with my reiserfs partitions. 
> The data is only copied from the partitions backup counterpart that was 
> freshly created. There was KDE up during copying.
> First partition: 1958788 kB (my system)
> ---------------------------------------
Hm. Only 2 gigabytes... ;)

> notail:
> real    5m16.853s user    0m1.660s sys     0m53.090s
> notail+iicache:
> real    4m55.041s user    0m1.820s sys     0m52.650s

Ok, so for this partition there are actually some good times.

> notail+noborder+iicache:
> real    5m32.181s user    0m1.260s sys     0m48.590s

Noborder is evil right now, don't use it if you are not sure what are you
doing ;)

> Second partition: 4085252 kB (mostly archives)
> ------------------------------------------
> notail:
> real    11m3.024s  user    0m1.460s sys     1m26.780s
> notail+iicache:
> real    11m13.734s user    0m1.620s sys     1m26.650s

Hm, things are not so good here.

Too bad you do not have any file-size distribution statistics.

> Yes, I have differences. But I think they're not the ones we wanted to 
> see?! I double checked the lines I pasted to this email ;-)

I believe right now we have better patches to speed-up things both on read
(IDE only) and write that we are testing mostly locally.
iicache only speedups some reads on large files (I have not looked at
new iicache write support yet, so my observations are only can be applied 
up to iicache-9 patch)

Bye,
    Oleg

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Manuel Krause]

On 04/08/2002 02:02 PM, Oleg Drokin wrote:

> Hello!
> 
> On Mon, Apr 08, 2002 at 01:44:48PM +0200, Manuel Krause wrote:
> 
> 
[...]
> Noborder is evil right now, don't use it if you are not sure what are you
> doing ;)


O.k., _then_ I should not use it! :-))

[...]
> Hm, things are not so good here.
> 
> Too bad you do not have any file-size distribution statistics.


Is there an easy way to get this info from the filesystems?

>>Yes, I have differences. But I think they're not the ones we wanted to 
>>see?! I double checked the lines I pasted to this email ;-)
>>
> 
> I believe right now we have better patches to speed-up things both on read
> (IDE only) and write that we are testing mostly locally.


Could you make them available on the namesys server? ;-)


> iicache only speedups some reads on large files (I have not looked at
> new iicache write support yet, so my observations are only can be applied 
> up to iicache-9 patch)
> 
> Bye,
>     Oleg

Thanks,

Manuel

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Nikita Danilov]

Manuel Krause writes:
 > On 04/08/2002 02:02 PM, Oleg Drokin wrote:
 > 
 > > Hello!
 > > 
 > > On Mon, Apr 08, 2002 at 01:44:48PM +0200, Manuel Krause wrote:
 > > 
 > > 
 > [...]
 > > Noborder is evil right now, don't use it if you are not sure what are you
 > > doing ;)
 > 
 > 
 > O.k., _then_ I should not use it! :-))
 > 
 > [...]
 > > Hm, things are not so good here.
 > > 
 > > Too bad you do not have any file-size distribution statistics.
 > 
 > 
 > Is there an easy way to get this info from the filesystems?
 > 

find / -type f -printf "%s\n" | sort -n | uniq -c | mail Green@Namesys.COM

or, even 

( 
	echo 'set terminal postscript;'
	echo 'clear;'
	echo 'plot [0:1000] [0:1000] "-" using 2 :1'
	find / -type f -printf "%k\n" | \
	sort -n | \
	uniq -c 
) | gnuplot | gv -landscape -


 > >>Yes, I have differences. But I think they're not the ones we wanted to 
 > >>see?! I double checked the lines I pasted to this email ;-)
 > >>
 > > 
 > > I believe right now we have better patches to speed-up things both on read
 > > (IDE only) and write that we are testing mostly locally.
 > 
 > 
 > Could you make them available on the namesys server? ;-)
 > 
 > 
 > > iicache only speedups some reads on large files (I have not looked at
 > > new iicache write support yet, so my observations are only can be applied 
 > > up to iicache-9 patch)
 > > 
 > > Bye,
 > >     Oleg
 > 
 > Thanks,
 > 
 > Manuel
 > 

Nikita.

 > 
 > 

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Oleg Drokin]

Hello!

On Mon, Apr 08, 2002 at 02:45:58PM +0200, Manuel Krause wrote:

> >Hm, things are not so good here.
> >Too bad you do not have any file-size distribution statistics.
> Is there an easy way to get this info from the filesystems?

Nikita gave you two examples ;)

> >>Yes, I have differences. But I think they're not the ones we wanted to 
> >>see?! I double checked the lines I pasted to this email ;-)
> >I believe right now we have better patches to speed-up things both on read
> >(IDE only) and write that we are testing mostly locally.
> Could you make them available on the namesys server? ;-)

As I said, we are still testing this stuff and do we do not want for someone to
accidentally apply the patches that were not tested enough.
I can send these to you if you want to perform some benchamrks, though.

Bye,
    Oleg

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Hans Reiser]

Oleg Drokin wrote:

>Hello!
>
>On Mon, Apr 08, 2002 at 02:45:58PM +0200, Manuel Krause wrote:
>
>>>Hm, things are not so good here.
>>>Too bad you do not have any file-size distribution statistics.
>>>
>>Is there an easy way to get this info from the filesystems?
>>
>
>Nikita gave you two examples ;)
>
>>>>Yes, I have differences. But I think they're not the ones we wanted to 
>>>>see?! I double checked the lines I pasted to this email ;-)
>>>>
>>>I believe right now we have better patches to speed-up things both on read
>>>(IDE only) and write that we are testing mostly locally.
>>>
>>Could you make them available on the namesys server? ;-)
>>
>
>As I said, we are still testing this stuff and do we do not want for someone to
>accidentally apply the patches that were not tested enough.
>I can send these to you if you want to perform some benchamrks, though.
>
>Bye,
>    Oleg
>
>
I'd like you to post the new code to the mailing list, and encourage 
people to play around with it (without using it for real data).

We need more people benchmarking it so that we can get some results and 
some discussions.

Hans

Re: Question on IICACHE and mount info (2.4.19-pre5)

[Yury Yu. Rupasov]

Manuel Krause wrote:

> Hi!
>
> Is there only the addidional mount option "iicache" with this patch:
> ftp://ftp.namesys.com/pub/reiserfs-for-2.4/testing/linux-2.4.19p4-iicache-new-5.patch
> ?
>
> There were some more in the early versions of iicache, IIRC.

Yes, you are right. The additional optimizing options removed for now.

>
>
> But at the moment it works very well with 2.4.19-pre5 +
> rml.preempt+lock-break + aa.vm-33 + some other aa patches and some akpm
> ones.
>
> When I put ",iicache" to the fstab lines of my reiserfs partitions I
> don't get a response from
>
> # cat "/proc/fs/reiserfs/ide0(3,11)/super"
> state:  REISERFS_VALID_FS
> mount options:  BORDER NO_TAILS LOG
> ...
>
> # cat "/proc/fs/reiserfs/ide0(3,7)/super"
> state:  REISERFS_VALID_FS
> mount options:  BORDER NO_TAILS LOG
> ...
>
> # cat /proc/mounts
> /dev/root / reiserfs rw,noatime 0 0
> /dev/hda11 /mnt/beta reiserfs rw,noatime,nosuid,nodev 0 0
> ...
>
> only # mount says:
> /dev/hda7 on / type reiserfs (rw,noatime,notail,iicache)
> ...
> /dev/hda11 on /mnt/beta type reiserfs
> (rw,nosuid,nodev,noatime,notail,iicache)
> ...
>
> Seems like some things have gone with the latest patches which worked
> before? It's o.k., of course, if "iicache" doesn't show up for now, I
> know it's for testing only (yes, it would be very nice!). But the
> "notail" thing worked before!

The problem is connected with broken proc_info here.

Here is iicache-new-10 patch, where proc_info problem is fixed :
ftp://ftp.namesys.com/pub/reiserfs-for-2.4/testing/linux-2.4.19p5-iicache-new-10.patch

Thanks,
Yura.