Hi
Following is the list of messages that I received for checking
and increasing the metadata extents.
Thank you.
Have a nice day.
Regards
____________________________________________________________________________
Chua Koon Teck
Singapore Telecom
Comcentre
SingNet
email=koonteck_at_singnet.com.sg
URL="
http://www.singnet.com.sg/~koonteck"
____________________________________________________________________________
-----BEGIN PGP PUBLIC KEY BLOCK-----
Version: 2.6.i
mQCNAjCz3oAAAAEEALaaeL0E8AJ6NkL2XZI/BLn5UH9W2oy+iwXf+8QolN1NlRuP
EYYgqt9s5pnmR5t1jpl+OWNzYLGiovsEUm8lgtlXrVFtI+SpsJTrA+tvpyRL0UNu
bAVeMZTc2g9loHF5zab0AIG7QUrTu88oUJjDHMThpxdFLzJ0CKFv0KQasFflAAUR
tChDaHVhIEtvb24gVGVjayA8a29vbnRlY2tAc2luZ25ldC5jb20uc2c+
=4XZx
-----END PGP PUBLIC KEY BLOCK-----
---------- Forwarded message ----------
Date: Sat, 04 May 1996 09:54:37 -0400
From: Steve Hancock <hancock_at_alf.dec.com>
To: Chua Koon Teck <koonteck_at_singnet.com.sg>
Cc: hancock_at_decatl.alf.dec.com
Subject: Re: increasing the metadata extent
Chua Koon Teck wrote:
>
> Hi
>
> I have a domain that is shown in the 'df -k' as :
>
> acct_domain#u0 14540208 1968670 4812352 29% /u0
> acct_domain#u1 14540208 7416487 4812352 61% /u1
>
> It is only 61% full. But I have been receiving "No space left on device"
> error messages when I use the adduser to add new user whose home directory
> is in the / u1 directory. I have about 30,000 home directories in the /u1
> dir.
>
> I am suspecting that the acct_domain has run out of space in the metadata
> table.
I suspect the same thing. To know for sure, use the following commands (assuming
you have only one volume in you AdvFS domain):
# cd /u0/.tags
# showfile -x M-6 (This shows how fragmented the metadata is in volume 1)
Here's an example:
Filesystem 1024-blocks Used Avail Capacity Iused Ifree %Iuse
d Mounted on
usr_domain#usr 11375104 8303152 2509248 77% 836419 9439296 8%
/usr
Id Vol PgSz Pages XtntType Segs SegSz Log Perf File
fffffffa.0000 1 16 61698 simple ** ** ftx 98% M-6
extentMap: 1
pageOff pageCnt vol volBlock blockCnt
0 2 1 32 32
2 128 1 12304 2048
130 128 1 306432 2048
258 128 1 341088 2048
386 128 1 375728 2048
514 128 1 410368 2048
642 128 1 445152 2048
770 128 1 479776 2048
898 128 1 514416 2048
1026 128 1 548304 2048
1154 128 1 582944 2048
[Snip...]
61604 34 1 14804432 544
61638 34 1 14923792 544
61672 11 1 2113296 176
61683 9 1 2114912 144
61692 5 1 2115232 80
61697 1 1 2115328 16
extentCnt: 683
Once you get to 683, you're done. The maximum number of metadata extents per volume
is 683.
> The previous setting for the metadata extents is 128. How do I
> increase the metadata extents of the acct_domain when the system is on
> line ?
The best way to fix this while the volume is on line is to add another volume to
the domain (using addvol), and then remove the volume that is fragmented (using
rmvol). Then, the files will automatically be migrated from the volume being removed
to the new volume and the metadata will be made contiguous at the same time. Then,
if you want, you can do the same procedure in reverse to make the volume back to the
way that it was.
if you do not have an extra disk, you must vdump/vrestore the domain (remake the domain
and filesets in between), which can be a time consuming process. I recommend the former
process if you have the spare disk.
> Is there any performance issue if I will to increase it to 1024 ?
>
There is no performance issue I am aware of. What increasing the value does is to
increase the default number of file system blocks the metadata will grow by, when
it needs to extend itself. What this implies, is that the metadata should have to
extend iteself less frequently because it grows by a larger value each time. The 1024
you mention here is a good number, IMHO, because very large extent values will tend
to become fragmented more quickly than smaller values. In other words, since the
file system will attempt to grab contiguous blocks on metadata extent size, but if
it can't due to file system fragmentation, it will get only the best available
contiguous space. That means you are getting less extent growth than you expect, and
can run out of your 683 extents much quicker. This leads to schemes like preallocation
of metadata extents, if you know you will have a problem. The only way to do that
with current released versions of Digital UNIX is to write a script or program
that creates millions of zero-length files and then removes them. In 4.0, there will
be a switch (-p) on the mkfdmn and addvol commands to preallocate as many extents as
you want (up to 683 per volume).
I could spend more time on this, but I think I answered the questions. If you have
additional questions about this, then e-mail me directly and I will try to answer
them. Please post a summary of all the responses to the osf-managers mailing list.
Regards,
Steve
--
| Steve Hancock | Digital UNIX Support |
| m/s ALF1-3/P26 | hancock_at_alf.dec.com |
| 5555 Windward Pkwy West | |
| Alpharetta, Ga 30201-7407 | #include <std/disclaimer.h> |
>From alan_at_nabeth.cxo.dec.com Tue May 7 13:58:28 1996
Date: Sat, 4 May 1996 00:21:44 -0600
From: "Alan Rollow - Dr. File System's Home for Wayward Inodes."
<alan_at_nabeth.cxo.dec.com>
To: koonteck_at_singnet.com.sg
Subject: Re: increasing the metadata extent
Backup the data, re-create the domain using the -x option to
specify a larger metadata extent. Preallocate and then remove
sufficient files for the needs of the application. Restore
from backup.
Alternately, it could be that the available space is badly
fragmented and that's why another extent couldn't be allocated.
If you have the AdvFS Utilities, run defragment.
Also, if you have one, consider adding another volume using
a more suitable extent size. Then remove the original volume
and wait for the files to migrate.
The extent size can only be set, when the domain is created
or a volume added.
Received on Tue May 07 1996 - 11:39:47 NZST