Discussion:
Arkeia: The Future of Arkeia, File Systems, and inodes
Matthew E.Porter
2007-03-29 16:39:53 UTC
Permalink
Arkeia Users:
I am starting this thread in hopes that it will serve as an open
dialogue between Arkeia and its community. It is not meant to spread
FUD. It is focused around what I see as the upcoming future problems
- related to Arkeia and its use of file systems and inodes.

A little background about Arkeia's internal data structure is
needed to fully analyze and appreciate the situation. Arkeia server
stores a series of information in the /opt/arkeia/server/dbase
directory. This directory contains a number of sub-directories, the
most critical to this conversation is the o3dbtree directory. The
o3dbtree directory is the beginning root of a hierarchical layout for
every server/device/appliance currently being backed up by the Arkeia
server.

Examining this directory is an interesting exercise. One sees a
directory structure that lists each system being backed up. Each
backed up system has a directory structure that mimics the directory
structure actually on the system as it was backed up. In essence,
imagine a tree for every system being backed up that contains only
the directories of that system with a series of marker files (the
o3_cpnt and o3_cpnt.lck files).

This structure is intuitive and easy to understand. There is
minimal black magic. The downside is that Arkeia requires 4 inodes
per directory (remember, these are the directories on the backed up
client): 2 for the directory itself, 1 for o3cpnt file, and 1 for the
o3cpnt.lck lock file. The downside is that you need a file system
that can handle this. The official stance from Arkeia is "In most
cases, if the disk space is enough to hold the database, then the i-
node number is enough. But, if you have a server which contains lot
of directories with few files you may have to change the i-node
configuration for your index database." (http://www.arkeia.com/
support/faq/networkbackup.php)

I disagree and see a significant issue approaching with Arkeia.
The amount of inodes has a limit on most major file systems. As
Linux is the recommended platform for deployment of Arkeia (server),
the issue is even more direct with the recent happenings in the file
system space. For years, Arkeia has recommended Reiserfs as the
solution to this problem. Unfortunately, the future Reiserfs is in a
relatively unknown and looming state. Consider the following:
- SuSE Linux has dropped ReiserFS as the default file system. See
http://linux.wordpress.com/2006/09/27/suse-102-ditching-reiserfs-as-
it-default-fs/ for a lengthy discussion
- Reiser 3 is in maintenance mode
- Reiser 4 does not appear that it will gain acceptance into the kernel
- The future of Hans Reiser and his company are in doubt given the
recent problems. See http://kcbs.com/pages/319259.php?
contentType=4&contentId=386120 if you have no clue why.

These are issues that will begin impacting more and more Arkeia
users as drives expand in size. Arkeia's competitors are noting that
this will impact the near term usage of Arkeia's users. With
consumer PCs shipping with drives as large as 1 TB (http://
tinyurl.com/2zdaj6), how can this not impact Arkeia deployments?

The question pops up "Doesn't Arkeia 6 solve these problems with
the media server deployment capabilities?". I do not fully know the
answer but believe that it does not. The media servers are
responsible for storage of the backed up data; however, the central
backup server still is responsible for the index directory (the
location of the o3dbtree structure). If my understanding is correct,
the ability for Arkeia to backup more data has been increased, and
this could actually cause more users to hit this limit.

(This skips over the current problems with this structure. For
example, it takes approximately 40 hours for an rsync to run on our /
opt/arkeia/server directory which has 42 GB of data in it. The
server is utilizing a RAID10 configuration with 15k Serial Attached
SCSI drives. It is nearly *impossible* for to backup this directory
through traditional methods.)

This was not mean to portray the situation as doom and gloom.
This does *NOT* mean people should run out and abandon Arkeia. In
reality, I am hoping for quite the opposite. I am hoping someone out
there has an answer to this. At bare minimum, I am hoping this will
open a dialogue with Arkeia. As a customer paying annual support, I
am hoping to hear that Arkeia is looking to the future and these issues.

Arkeia is a great piece of software, and our company has a
significant investment in it. We have developed custom applications
to provide features that we need, including integration with our
management system. We are happy Arkeia users. I want this statement
to still be true in 2 to 5 years.


Cheers,
Matthew

---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Frank Bures
2007-03-29 19:09:38 UTC
Permalink
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Post by Matthew E.Porter
(This skips over the current problems with this structure. For
example, it takes approximately 40 hours for an rsync to run on our /
opt/arkeia/server directory which has 42 GB of data in it. The
server is utilizing a RAID10 configuration with 15k Serial Attached
SCSI drives. It is nearly *impossible* for to backup this directory
through traditional methods.)
Wow, that's quite a number.
I run ARKEIA to backup some 16 machines. Total backup is worth some 150GB.
Daily incremental backups some 35GB. My /opt/arkeia/server directory is 2GB
in size.
I am backing up to disks (RAID5), currently the overall size of the backup is
around 3TB.

I backup my /opt/arkeia/server/dbase by simply 'tar -cvfz', putting it on a
separate mirrored disk and then rotating it over 7 days period. Current
tar.gz size is some 55MB. To tar /opt/arkeia/server/dbase takes some 20
minutes.

I find your size of /opt/arkeia/server quite disturbing. Have you ever run
'/opt/arkeia/bin/arkdbchk -p' ?

Cheers
Frank


Frank Bures, Dept. of Chemistry, University of Toronto, M5S 3H6
fbures-q9rj9x3S2gwTRQv+KTR+***@public.gmane.org
http://www.chem.utoronto.ca
PGP public key: http://pgp.mit.edu:11371/pks/lookup?op=index&search=Frank+Bures

Please, reply to the above address only. <lisfrank> does not accept replies.
-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 5.0 OS/2 for non-commercial use
Comment: PGP 5.0 for OS/2
Charset: cp850

wj8DBQFGDADiih0Xdz1+w+wRAn6jAKCixgH25QBnMGnDpYQCulMEzkUSPwCfQD1/
XstcIbAPCTO60Oy6TyPZgBM=
=LLPl
-----END PGP SIGNATURE-----


___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Matthew E. Porter
2007-03-29 19:53:16 UTC
Permalink
Post by Frank Bures
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Post by Matthew E.Porter
(This skips over the current problems with this structure. For
example, it takes approximately 40 hours for an rsync to run on our /
opt/arkeia/server directory which has 42 GB of data in it. The
server is utilizing a RAID10 configuration with 15k Serial Attached
SCSI drives. It is nearly *impossible* for to backup this directory
through traditional methods.)
Wow, that's quite a number.
I run ARKEIA to backup some 16 machines. Total backup is worth some 150GB.
Daily incremental backups some 35GB. My /opt/arkeia/server
directory is 2GB
in size.
I am backing up to disks (RAID5), currently the overall size of the backup is
around 3TB.
I backup my /opt/arkeia/server/dbase by simply 'tar -cvfz', putting it on a
separate mirrored disk and then rotating it over 7 days period.
Current
tar.gz size is some 55MB. To tar /opt/arkeia/server/dbase takes some 20
minutes.
I find your size of /opt/arkeia/server quite disturbing. Have you ever run
'/opt/arkeia/bin/arkdbchk -p' ?
Frank:
Thank you for the concern. We back up approximately 65 TB (*not*
a typo) per month.

Yep, we run the purge every few weeks. The purge also takes
approximately

Regardless of our size of backups, the issue truly is related to
file system constraints due to the number of client directories. In
a world where there are now media servers, this problem will only be
compounded.



Cheers,
Matthew

---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Graeme Hinchliffe
2007-03-29 20:46:49 UTC
Permalink
Post by Matthew E. Porter
Thank you for the concern. We back up approximately 65 TB (*not*
a typo) per month.
Yep, we run the purge every few weeks. The purge also takes
approximately
Regardless of our size of backups, the issue truly is related to
file system constraints due to the number of client directories.
In a world where there are now media servers, this problem will
only be compounded.
We don't backup that much data, but as we do backup webservers which
host 10,000's of websites as well as all servers our server directory
is also a bit of a huge state. We have seen problems when it comes
to restorations as it takes an age to trawl through the data
structure there, this has been seen with the disks running at 100% of
their ability whilst arkeia 'thinks'.

It is a concern as I would like to backup some servers which hold a
LOT of small rapidly changing files, and with this current structure
I can forsee problems.

I switched to ReiserFS sue to it's INODE less operation and seemingly
good handling of small files, though with recent FS tests I may
rethink this to XFS which seems much quicker.




-----
Graeme Hinchliffe (BSc)
Senior Systems Engineer
Zen Internet (http://www.zen.co.uk/)

Direct: 0845 058 9074
Main : 0845 058 9000
Fax : 0845 058 9005
Matthew E. Porter
2007-03-29 22:48:01 UTC
Permalink
Post by Graeme Hinchliffe
Post by Matthew E. Porter
Thank you for the concern. We back up approximately 65 TB
(*not* a typo) per month.
Yep, we run the purge every few weeks. The purge also takes
approximately
Regardless of our size of backups, the issue truly is related to
file system constraints due to the number of client directories.
In a world where there are now media servers, this problem will
only be compounded.
We don't backup that much data, but as we do backup webservers
which host 10,000's of websites as well as all servers our server
directory is also a bit of a huge state. We have seen problems
when it comes to restorations as it takes an age to trawl through
the data structure there, this has been seen with the disks running
at 100% of their ability whilst arkeia 'thinks'.
It is a concern as I would like to backup some servers which hold a
LOT of small rapidly changing files, and with this current
structure I can forsee problems.
I switched to ReiserFS sue to it's INODE less operation and
seemingly good handling of small files, though with recent FS tests
I may rethink this to XFS which seems much quicker.
Graeme and/or Arkeia Folks:
Have you done any testing of the dbase directory on XFS? Are
there any known problems?

Are these tests ones you have run or publicly available? I would
love to see them.


Cheers,
Matthew


---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Graeme Hinchliffe
2007-03-30 07:13:19 UTC
Permalink
Post by Matthew E. Porter
Have you done any testing of the dbase directory on XFS? Are
there any known problems?
Are these tests ones you have run or publicly available? I would
love to see them.
The tests I ran were using the postfix disk test utility freely
released by netapp. Testing for file access of small files on an XFS
platform it was nearly twice as fast as a reiserfs platform and twice
as fast as ext3. The test I run were for another project but proved
useful elsewhere too.

Just had a dig round and my test results were as follows :

Current configuration is:
The base number of files is 10000
Transactions: 10000
Files range between 500 bytes and 48.83 kilobytes in size
Working directory:
/test1 (weight=1)
1000 subdirectories will be used
Block sizes are: read=512 bytes, write=512 bytes
Biases are: read/append=5, create/delete=5
Using Unix buffered file I/O
Random number generator seed is 42
Report format is verbose.



XFS
Time:
333 seconds total
151 seconds of transactions (66 per second)

Files:
14957 created (44 per second)
Creation alone: 10000 files (166 per second)
Mixed with transactions: 4957 files (32 per second)
5009 read (33 per second)
4991 appended (33 per second)
14957 deleted (44 per second)
Deletion alone: 9914 files (81 per second)
Mixed with transactions: 5043 files (33 per second)

Data:
131.21 megabytes read (403.49 kilobytes per second)
413.44 megabytes written (1.24 megabytes per second)

Reiser
Time:
544 seconds total
388 seconds of transactions (25 per second)

Files:
14957 created (27 per second)
Creation alone: 10000 files (101 per second)
Mixed with transactions: 4957 files (12 per second)
5009 read (12 per second)
4991 appended (12 per second)
14957 deleted (27 per second)
Deletion alone: 9914 files (173 per second)
Mixed with transactions: 5043 files (12 per second)

Data:
131.21 megabytes read (246.99 kilobytes per second)
413.44 megabytes written (778.24 kilobytes per second)

ext3
Time:
603 seconds total
302 seconds of transactions (33 per second)

Files:
14957 created (24 per second)
Creation alone: 10000 files (48 per second)
Mixed with transactions: 4957 files (16 per second)
5009 read (16 per second)
4991 appended (16 per second)
14957 deleted (24 per second)
Deletion alone: 9914 files (105 per second)
Mixed with transactions: 5043 files (16 per second)

Data:
131.21 megabytes read (222.82 kilobytes per second)
413.44 megabytes written (702.09 kilobytes per second)


HTH

Graeme
stephane.purnelle-r9t4bu4u4bmzQB+
2007-03-30 07:47:18 UTC
Permalink
Hi,

My dbase directory is on XFS and I have no problem.

-----------------------------------
Stéphane PURNELLE stephane.purnelle-r9t4bu4u4bmzQB+***@public.gmane.org
Service Informatique Corman S.A. Tel : 00 32 087/342467



Graeme Hinchliffe <graeme.hinchliffe-3Sf606SdvB6RJKlD0jw6FVpr/1R2p/***@public.gmane.org>
Envoyé par : owner-arkeia-userlist-xIg/***@public.gmane.org
30/03/2007 09:13
Veuillez répondre à
arkeia-userlist-xIg/***@public.gmane.org


A
arkeia-userlist-xIg/***@public.gmane.org
cc

Objet
Re: [arkeia-userlist] Arkeia: The Future of Arkeia, File Systems, and
inodes







On 29 Mar 2007, at 23:48, Matthew E. Porter wrote:

Graeme and/or Arkeia Folks:
Have you done any testing of the dbase directory on XFS? Are there any
known problems?

Are these tests ones you have run or publicly available? I would love
to see them.

The tests I ran were using the postfix disk test utility freely released
by netapp. Testing for file access of small files on an XFS platform it
was nearly twice as fast as a reiserfs platform and twice as fast as
ext3. The test I run were for another project but proved useful elsewhere
too.

Just had a dig round and my test results were as follows :

Current configuration is:
The base number of files is 10000
Transactions: 10000
Files range between 500 bytes and 48.83 kilobytes in size
Working directory:
/test1 (weight=1)
1000 subdirectories will be used
Block sizes are: read=512 bytes, write=512 bytes
Biases are: read/append=5, create/delete=5
Using Unix buffered file I/O
Random number generator seed is 42
Report format is verbose.



XFS
Time:
333 seconds total
151 seconds of transactions (66 per second)

Files:
14957 created (44 per second)
Creation alone: 10000 files (166 per second)
Mixed with transactions: 4957 files (32 per second)
5009 read (33 per second)
4991 appended (33 per second)
14957 deleted (44 per second)
Deletion alone: 9914 files (81 per second)
Mixed with transactions: 5043 files (33 per second)

Data:
131.21 megabytes read (403.49 kilobytes per second)
413.44 megabytes written (1.24 megabytes per second)

Reiser
Time:
544 seconds total
388 seconds of transactions (25 per second)

Files:
14957 created (27 per second)
Creation alone: 10000 files (101 per second)
Mixed with transactions: 4957 files (12 per second)
5009 read (12 per second)
4991 appended (12 per second)
14957 deleted (27 per second)
Deletion alone: 9914 files (173 per second)
Mixed with transactions: 5043 files (12 per second)

Data:
131.21 megabytes read (246.99 kilobytes per second)
413.44 megabytes written (778.24 kilobytes per second)

ext3
Time:
603 seconds total
302 seconds of transactions (33 per second)

Files:
14957 created (24 per second)
Creation alone: 10000 files (48 per second)
Mixed with transactions: 4957 files (16 per second)
5009 read (16 per second)
4991 appended (16 per second)
14957 deleted (24 per second)
Deletion alone: 9914 files (105 per second)
Mixed with transactions: 5043 files (16 per second)

Data:
131.21 megabytes read (222.82 kilobytes per second)
413.44 megabytes written (702.09 kilobytes per second)


HTH

Graeme
harald glanzer
2007-03-30 13:35:35 UTC
Permalink
Post by Graeme Hinchliffe
Post by stephane.purnelle-r9t4bu4u4bmzQB+
Have you done any testing of the dbase directory on XFS? Are there
any known problems?
Are these tests ones you have run or publicly available? I would
love to see them.
The tests I ran were using the postfix disk test utility freely
released by netapp. Testing for file access of small files on an XFS
platform it was nearly twice as fast as a reiserfs platform and twice
as fast as ext3. The test I run were for another project but proved
useful elsewhere too.
The base number of files is 10000
Transactions: 10000
Files range between 500 bytes and 48.83 kilobytes in size
/test1 (weight=1)
1000 subdirectories will be used
Block sizes are: read=512 bytes, write=512 bytes
Biases are: read/append=5, create/delete=5
Using Unix buffered file I/O
Random number generator seed is 42
Report format is verbose.
XFS
333 seconds total
151 seconds of transactions (66 per second)
14957 created (44 per second)
Creation alone: 10000 files (166 per second)
Mixed with transactions: 4957 files (32 per second)
5009 read (33 per second)
4991 appended (33 per second)
14957 deleted (44 per second)
Deletion alone: 9914 files (81 per second)
Mixed with transactions: 5043 files (33 per second)
131.21 megabytes read (403.49 kilobytes per second)
413.44 megabytes written (1.24 megabytes per second)
Reiser
544 seconds total
388 seconds of transactions (25 per second)
14957 created (27 per second)
Creation alone: 10000 files (101 per second)
Mixed with transactions: 4957 files (12 per second)
5009 read (12 per second)
4991 appended (12 per second)
14957 deleted (27 per second)
Deletion alone: 9914 files (173 per second)
Mixed with transactions: 5043 files (12 per second)
131.21 megabytes read (246.99 kilobytes per second)
413.44 megabytes written (778.24 kilobytes per second)
ext3
603 seconds total
302 seconds of transactions (33 per second)
14957 created (24 per second)
Creation alone: 10000 files (48 per second)
Mixed with transactions: 4957 files (16 per second)
5009 read (16 per second)
4991 appended (16 per second)
14957 deleted (24 per second)
Deletion alone: 9914 files (105 per second)
Mixed with transactions: 5043 files (16 per second)
131.21 megabytes read (222.82 kilobytes per second)
413.44 megabytes written (702.09 kilobytes per second)
HTH
Graeme
hello,

interesting to hear about this discussion - the arkeia-index 'database'
is a problem in our company too.
we recently had the problem of running out of inodes on the
ext3-partition holding the arkeia-installation
on our backupserver.
we now switched to XFS, and there are no problems yet. a big
advantage(beside the speed) is that XFS handles
the inodes dynamically, so you should not run out of inodes, as long you
have free diskspace.

greetz
harald glanzer
vienna

___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
David Guntner
2007-03-30 14:44:02 UTC
Permalink
I've tried everything I can think of to unsubscribe from this list, but
nothing seems to work and the messages keep coming. Can someone please
point me at how to unsubscribe?

--Dave
Matthew E. Porter
2007-03-30 15:03:58 UTC
Permalink
Dave:
I don't work for Arkeia, but found this for you - http://
www.arkeia.com/support/usermailinglist/index.php .

Did you email that person?


Cheers,
Matthew

---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Post by David Guntner
I've tried everything I can think of to unsubscribe from this list, but
nothing seems to work and the messages keep coming. Can someone please
point me at how to unsubscribe?
--Dave
Nick Moulsdale
2007-04-03 09:11:40 UTC
Permalink
I'd like to recycle my tapes as Arkeia has got itself in a muddle, asking
for tapes out of sequence and refusing to accept the ones I want it to use.

BUT last time I recycled I lost all the history for all the tapes.

Is there another way that tells the tapes to allow to be re-used in MY order
but doesn't lose the history? We do a complete back up every weekday (as set
by calendar) using 2 Tapes.

Any help gratefully received.

Nick


N.G. Moulsdale FCA
Group Finance Director
Ebi Manufacturing Co Ltd
Sandhill House
82 Meanwood Road
Leeds LS7 2RE
United Kingdom
* 44-(0)113-243-2448
Fax 44-(0)113-243-0504
* Nick-***@public.gmane.org





NOTE: The information contained in this Email is intended for the named recipient(s) only. It may also be privileged and confidential. If you are not an intended recipient, you must not copy, distribute or take any action in reliance upon it. No warranties or assurances are made in relation to the safety and content of this e-mail and any attachments. No liability is accepted for any consequences arising from it.



-------------------------------------------------------------------------------------------------------------
EBI Manufacturing Co Ltd 82 Meanwood Road, Leeds, LS7 2RE, UK
Co No 1447807 VAT GB 613 3570 65
Philippe Breider
2007-04-10 07:20:07 UTC
Permalink
Hi Nick,
Post by Nick Moulsdale
BUT last time I recycled I lost all the history for all the tapes.
As recycling a tape enables to overwrite its content it's normal that
the history of the data stored on it is also deleted .
Post by Nick Moulsdale
Is there another way that tells the tapes to allow to be re-used in MY
order but doesn't lose the history? We do a complete back up every
weekday (as set by calendar) using 2 Tapes.
No

Regards,

Philippe
--
Philippe Breider, EMEA Pre-sales Engineer
Tel: 33 (0)1 48 10 89 79 - Fax: 33 (0)1 48 10 89 90
Arkeia Corp. - Back up your data, Recover your business

___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Carsten Schaub
2007-07-31 09:30:53 UTC
Permalink
Post by stephane.purnelle-r9t4bu4u4bmzQB+
Post by Graeme Hinchliffe
Post by Matthew E. Porter
Thank you for the concern. We back up approximately 65 TB
(*not* a typo) per month.
Yep, we run the purge every few weeks. The purge also takes
approximately
Regardless of our size of backups, the issue truly is related to
file system constraints due to the number of client directories.
In a world where there are now media servers, this problem will
only be compounded.
We don't backup that much data, but as we do backup webservers which
host 10,000's of websites as well as all servers our server
directory is also a bit of a huge state. We have seen problems when
it comes to restorations as it takes an age to trawl through the
data structure there, this has been seen with the disks running at
100% of their ability whilst arkeia 'thinks'.
It is a concern as I would like to backup some servers which hold a
LOT of small rapidly changing files, and with this current structure
I can forsee problems.
I switched to ReiserFS sue to it's INODE less operation and
seemingly good handling of small files, though with recent FS tests
I may rethink this to XFS which seems much quicker.
Have you done any testing of the dbase directory on XFS? Are there
any known problems?
I have switched to XFS for exactly that reason. It is much faster. The
backup before that switch took about 14 hours. Now with XFS underneath
it takes only 8 hours. Direct observed.

Carsten Schaub
Post by stephane.purnelle-r9t4bu4u4bmzQB+
Are these tests ones you have run or publicly available? I would
love to see them.
Cheers,
Matthew
---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
stephane.purnelle-r9t4bu4u4bmzQB+
2007-07-31 09:35:41 UTC
Permalink
We use XFS too, but actually we think to replace XFS with GFS, any advice
about GFS and arkeia .

Thanks

Stéphane Purnelle
SysAdmin

-----------------------------------
Stéphane PURNELLE stephane.purnelle-r9t4bu4u4bmzQB+***@public.gmane.org
Service Informatique Corman S.A. Tel : 00 32 087/342467
Post by Carsten Schaub
Post by stephane.purnelle-r9t4bu4u4bmzQB+
Post by Graeme Hinchliffe
Post by Matthew E. Porter
Thank you for the concern. We back up approximately 65 TB
(*not* a typo) per month.
Yep, we run the purge every few weeks. The purge also takes
approximately
Regardless of our size of backups, the issue truly is related to
file system constraints due to the number of client directories.
In a world where there are now media servers, this problem will
only be compounded.
We don't backup that much data, but as we do backup webservers which
host 10,000's of websites as well as all servers our server
directory is also a bit of a huge state. We have seen problems when
it comes to restorations as it takes an age to trawl through the
data structure there, this has been seen with the disks running at
100% of their ability whilst arkeia 'thinks'.
It is a concern as I would like to backup some servers which hold a
LOT of small rapidly changing files, and with this current structure
I can forsee problems.
I switched to ReiserFS sue to it's INODE less operation and
seemingly good handling of small files, though with recent FS tests
I may rethink this to XFS which seems much quicker.
Have you done any testing of the dbase directory on XFS? Are there
any known problems?
I have switched to XFS for exactly that reason. It is much faster. The
backup before that switch took about 14 hours. Now with XFS underneath
it takes only 8 hours. Direct observed.
Carsten Schaub
Post by stephane.purnelle-r9t4bu4u4bmzQB+
Are these tests ones you have run or publicly available? I would
love to see them.
Cheers,
Matthew
---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
If you wish to modify your subscription or if you want to subscribe
http://www.arkeia.com/community/
___________________________________________________________________
Frank Bures
2007-03-30 17:07:30 UTC
Permalink
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Post by Matthew E. Porter
We back up approximately 65 TB (*not*
a typo) per month.
Well, that explains it :-)


Frank Bures, Dept. of Chemistry, University of Toronto, M5S 3H6
fbures-q9rj9x3S2gwTRQv+KTR+***@public.gmane.org
http://www.chem.utoronto.ca
PGP public key: http://pgp.mit.edu:11371/pks/lookup?op=index&search=Frank+Bures

Please, reply to the above address only. <lisfrank> does not accept replies.
-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 5.0 OS/2 for non-commercial use
Comment: PGP 5.0 for OS/2
Charset: cp850

wj8DBQFGDTXCih0Xdz1+w+wRAhYKAKCc23i8SZO+0HMIwX9dEd+y5arG4wCg42R6
O33pSZFI2cR/mCWzUiKSxyw=
=vkq5
-----END PGP SIGNATURE-----


___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Oliver Schulze L.
2007-04-08 15:58:24 UTC
Permalink
Hi Matthew,
I agree with you. Also, there is no need to compare perfomance numbers since
the problem you are describing is by design not the best.

I hope Arkeia6 uses some kind of virtual filesystem or even better a
DataBase
for the dbase backend. Having a directory structure in a DB or a VFS
translate
just in characters with some structure, so it will save a lot of space
and inode.
And if you have a compressed VFS, well, thats will be just great ;)

HTH
Oliver
--
Oliver Schulze L.
My email after a captcha | http://tinymailto.com/oliverplanet
Home page | http://www.pla.net.py/home/oliver/

___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Matthew E. Porter
2007-04-08 16:31:31 UTC
Permalink
Oliver:
Thank you for your email. I am happy to see that people are
taking this seriously.

I am disappointed to report that this has *not* been resolved in
Arkeia 6. The data structure is exactly the same.

We have begun rolling out Arkeia 6 while maintaining our 5.3
system in parallel. Here is something disturbing - a "du -sh ."
command run on /opt/arkeia/server/dbase takes approximately 5.5
minutes on a dual CPU quad-core server with 4x146 GB drives running
in a RAID10 configuration. That directory is *only* 4.6 GB.

Has Arkeia commented on this at all? Is there any concern? I
recently heard that they have sold some large installations. How are
they planning to implement this?


Cheers,
Matthew

---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Post by Oliver Schulze L.
Hi Matthew,
I agree with you. Also, there is no need to compare perfomance
numbers since
the problem you are describing is by design not the best.
I hope Arkeia6 uses some kind of virtual filesystem or even better
a DataBase
for the dbase backend. Having a directory structure in a DB or a
VFS translate
just in characters with some structure, so it will save a lot of
space and inode.
And if you have a compressed VFS, well, thats will be just great ;)
HTH
Oliver
--
Oliver Schulze L.
My email after a captcha | http://tinymailto.com/oliverplanet
Home page | http://www.pla.net.py/home/oliver/
___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
If you wish to modify your subscription or if you want to subscribe
http://www.arkeia.com/community/
___________________________________________________________________
Frank Bures
2007-04-09 14:16:39 UTC
Permalink
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Post by Matthew E. Porter
Here is something disturbing - a "du -sh ."
command run on /opt/arkeia/server/dbase takes approximately 5.5
minutes on a dual CPU quad-core server with 4x146 GB drives running
in a RAID10 configuration. That directory is *only* 4.6 GB.
Just for comparison:
The same command over my 2GB directory takes 8 seconds.
The directory has approx 446k entries (ls -lR dbase | wc -l).
SW RAID-5 over 12 SATA disks. Athlon 4800+, RHEL4.

It seems that the problem starts with certain size of the directory (certain
number of entries).

Cheers
Frank


Frank Bures, Dept. of Chemistry, University of Toronto, M5S 3H6
fbures-q9rj9x3S2gwTRQv+KTR+***@public.gmane.org
http://www.chem.utoronto.ca
PGP public key: http://pgp.mit.edu:11371/pks/lookup?op=index&search=Frank+Bures

Please, reply to the above address only. <lisfrank> does not accept replies.
-----BEGIN PGP SIGNATURE-----
Version: PGPfreeware 5.0 OS/2 for non-commercial use
Comment: PGP 5.0 for OS/2
Charset: cp850

wj8DBQFGGjy3ih0Xdz1+w+wRAti8AKCm7ZDtzTDUZ5R6nXiDtSpPUlxD2wCZAVG3
YyvCXglzDFiktucpNgB8BM0=
=LhiH
-----END PGP SIGNATURE-----


___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Matthew E. Porter
2007-04-09 14:48:07 UTC
Permalink
Post by Frank Bures
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Post by Matthew E. Porter
Here is something disturbing - a "du -sh ."
command run on /opt/arkeia/server/dbase takes approximately 5.5
minutes on a dual CPU quad-core server with 4x146 GB drives running
in a RAID10 configuration. That directory is *only* 4.6 GB.
The same command over my 2GB directory takes 8 seconds.
The directory has approx 446k entries (ls -lR dbase | wc -l).
SW RAID-5 over 12 SATA disks. Athlon 4800+, RHEL4.
It seems that the problem starts with certain size of the directory (certain
number of entries).
I ran the same command - 6.1 million entries.


Cheers,
Matthew


---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Oliver Schulze L.
2007-04-10 23:41:15 UTC
Permalink
Hi Matthew,
thanks for your email.

Since ARK6 has the same issue, and waiting for an official reponse
solution could take long,
maybe you have aliviate your problems a litle with:
- using dir_index option in ext3
- using a 8GB solid state drive (Ram HardDisk)
- find some kind of SQL VFS, there was a couple of project in the past
but none actually
activelly developed

HTH
Oliver
Post by Matthew E. Porter
I am disappointed to report that this has *not* been resolved in
Arkeia 6. The data structure is exactly the same.
--
Oliver Schulze L.
My email after a captcha | http://tinymailto.com/oliverplanet
Home page | http://www.pla.net.py/home/oliver/

___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Matthew E. Porter
2007-12-14 04:46:36 UTC
Permalink
A few people have emailed to ask if anything ever changed and where we
were as a customer. At the end of the day, we outgrew Arkeia
specifically because of these issue. We are now approaching 80 TB of
backups monthly. We could no longer ignore or work around the
limitations of Arkeia. Arkeia is still a great product for certain
market segments. We just outgrew them. (They do need some
improvements like reporting and fixing email notifications which break
if the backup process dies.)

I do want to thank Sam and Jay from Arkeia support for all of their
help the past few years. It was truly a pleasure to work with you
both and helping us scale our Arkeia deployment to the level we were.
As someone who runs a company with a large support team and focuses on
support first, I think you are both phenomenal and have my utmost
respect.


Sincerely,
Matthew

---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Post by Matthew E. Porter
Thank you for your email. I am happy to see that people are
taking this seriously.
I am disappointed to report that this has *not* been resolved in
Arkeia 6. The data structure is exactly the same.
We have begun rolling out Arkeia 6 while maintaining our 5.3
system in parallel. Here is something disturbing - a "du -sh ."
command run on /opt/arkeia/server/dbase takes approximately 5.5
minutes on a dual CPU quad-core server with 4x146 GB drives running
in a RAID10 configuration. That directory is *only* 4.6 GB.
Has Arkeia commented on this at all? Is there any concern? I
recently heard that they have sold some large installations. How
are they planning to implement this?
Cheers,
Matthew
---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Post by Oliver Schulze L.
Hi Matthew,
I agree with you. Also, there is no need to compare perfomance numbers since
the problem you are describing is by design not the best.
I hope Arkeia6 uses some kind of virtual filesystem or even better
a DataBase
for the dbase backend. Having a directory structure in a DB or a
VFS translate
just in characters with some structure, so it will save a lot of
space and inode.
And if you have a compressed VFS, well, thats will be just great ;)
HTH
Oliver
--
Oliver Schulze L.
My email after a captcha | http://tinymailto.com/oliverplanet
Home page | http://www.pla.net.py/home/oliver/
___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
If you wish to modify your subscription or if you want to subscribe
http://www.arkeia.com/community/
___________________________________________________________________
___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Oliver Schulze L.
2007-06-01 04:32:56 UTC
Permalink
It seems that the list stopped working after this post :(
--
Oliver Schulze L.
My email after a captcha | http://tinymailto.com/oliverplanet
Home page | http://www.pla.net.py/home/oliver/

___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Arno Lehmann
2007-06-01 22:56:37 UTC
Permalink
Hi,
Post by Oliver Schulze L.
It seems that the list stopped working after this post :(
It seems that the list works but people are too shocked to write any
more ;-)

Arno
--
IT-Service Lehmann al-***@public.gmane.org
Arno Lehmann http://www.its-lehmann.de
___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Matthew E. Porter
2007-06-01 23:03:32 UTC
Permalink
Post by Arno Lehmann
Hi,
Post by Oliver Schulze L.
It seems that the list stopped working after this post :(
It seems that the list works but people are too shocked to write
any more ;-)
Actually, I submitted a support request a few weeks back about the
list being broken....

This topic is still relevant in my mind. FWIW, I hope to have a blog
post up shortly about how we scaled Arkeia to 50 TB on a single server.


Cheers,
Matthew

---
Matthew E. Porter
Contegix
Beyond Managed Hosting(r) for Your Enterprise
Oliver Schulze L.
2007-06-02 15:12:38 UTC
Permalink
Ooops,
I had an old list email address, here it is:

Old list address: User-List <arkeia-userlist-AHySt852JRvQT0dZR+***@public.gmane.org>
New list address: arkeia-userlist-xIg/***@public.gmane.org

HTH
Oliver
--
Oliver Schulze L.
My email after a captcha | http://tinymailto.com/oliverplanet
Home page | http://www.pla.net.py/home/oliver/

___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
d***@public.gmane.org
2007-12-14 05:14:05 UTC
Permalink
Bonjour,

J'ai quitté mes fonctions à l'ENGREF depuis le 30/11/2007.

Pour toute question d'ordre professionnel vous pouvez adresser vos demandes au Service Informatique de l'ENGREF Paris (service.informatique.paris-BGYPSvrStMVQFI55V6+***@public.gmane.org ou au 01.45.49.89.03).

Pour toute correspondance personnelle, vous pouvez m'écrire à l'adresse : anael.delorme-4gXcXVGLCurza9/***@public.gmane.org

Cordialement,

Anaël Delorme


___________________________________________________________________
Users of Arkeia can share their experiences and questions via this
mailing-list. Support programs are available to Arkeia commercial
version users: contact us at info-AHySt852JRvQT0dZR+***@public.gmane.org for more information.
If you wish to modify your subscription or if you want to subscribe
to other Arkeia mailing-lists, please, go to:
http://www.arkeia.com/community/
___________________________________________________________________
Loading...