Suggestions for new installation: hardware and backup

[davidedg]

Greetings.

We are going to deploy a Scalix installation as a replacement for another mail server from Visnetic.

Our new server hardware is a Hewlett-Packard Proliant DL-360 G4 with:
2 Xeon 3GHz cpus
2 Gigs RAM
2x160GB scsi disks (in hardware RAID1)

The existing user base is ~90 users with an occupation of ~512MB per mailbox.

We are "forced" to use this hardware as it has already been bought for this purpose.

We will use an Active Directory (2003 native mode + R2) domain with Scalix AD-schema extentions.
Linux Server OS will be a RedHat Enterprise 4 ES used solely for this purpose.

All scalix users will be premium users and accessing mailboxes mainly through Outlook (2000 or later) with the mapi connector. They will sometimes use external web access with SWA and with mobile phones.


My main concern is that, given the number of users (which, we were told, may double in 1-year time), the DISK SUBSYSTEM may be inadequate in speed, let alone for space requirements.

- May disk speed be an issue with 90 users @50GB of total message store ?

- And with 180 users ? Since the new 90 users will be on another (100Mbps-connected) site... will we be able to deploy another scalix server and "connect" it to the first scalix server... in the way Exchange works I mean...?

- Should we mount the scalix temp (~/temp ?) in a dedicated partition (maybe in-ram and/or mounted with kamikaze options) to optimize speed as I read somewhere on the forums ?

- - - -

Secondly, the backup task:

we'll be using LVM for snapshots with omsuspend and take a full daily backup (monday-friday) either directly to tape or by using disk-to-disk-to-tape stategy and during off-peak hours.

- Roughly, how much space should we leave as "free" in the volume group for the creation of the lvm snapshot ?

- Is there a suggested LVM extent-size (since scalix write many small files) or the 4MB default is ok ?

Our backup suite is Veritas Backup Exec with a VXA tape-library (10x1) attached.

- Can backups of the lvm snapshot be made through SAMBA/SMB or NFS ?
-> better with Linux server pushing data via rsync(?) to a SMB or NFS share on the Windows machine and then back it up (disk-to-disk-to-tape strategy)
or
-> Windows machine pulling data from a SAMBA or NFS share on the scalix server (that points to the snapshot) within the backup process (disk-to-tape)

The former requiring much more space but should be faster, the latter viceversa... but are SMB or NFS viable in any case?

- Any other suggestions/directions on this (or other) topic would be greatly appreciated... !!!


Please be as much verbose as you can because this is the first "big" installation we make.

Thank you and keep up the good job


--
Davide DG.

[Kris]

Greetings.

We are going to deploy a Scalix installation as a replacement for another mail server from Visnetic.

Our new server hardware is a Hewlett-Packard Proliant DL-360 G4 with:
2 Xeon 3GHz cpus
2 Gigs RAM
2x160GB scsi disks (in hardware RAID1)

The existing user base is ~90 users with an occupation of ~512MB per mailbox.

We are "forced" to use this hardware as it has already been bought for this purpose.

We will use an Active Directory (2003 native mode + R2) domain with Scalix AD-schema extentions.
Linux Server OS will be a RedHat Enterprise 4 ES used solely for this purpose.

All scalix users will be premium users and accessing mailboxes mainly through Outlook (2000 or later) with the mapi connector. They will sometimes use external web access with SWA and with mobile phones.


My main concern is that, given the number of users (which, we were told, may double in 1-year time), the DISK SUBSYSTEM may be inadequate in speed, let alone for space requirements.

- May disk speed be an issue with 90 users @50GB of total message store ?

- And with 180 users ? Since the new 90 users will be on another (100Mbps-connected) site... will we be able to deploy another scalix server and "connect" it to the first scalix server... in the way Exchange works I mean...?

- Should we mount the scalix temp (~/temp ?) in a dedicated partition (maybe in-ram and/or mounted with kamikaze options) to optimize speed as I read somewhere on the forums ?

- - - -

Secondly, the backup task:

we'll be using LVM for snapshots with omsuspend and take a full daily backup (monday-friday) either directly to tape or by using disk-to-disk-to-tape stategy and during off-peak hours.

- Roughly, how much space should we leave as "free" in the volume group for the creation of the lvm snapshot ?

- Is there a suggested LVM extent-size (since scalix write many small files) or the 4MB default is ok ?

Our backup suite is Veritas Backup Exec with a VXA tape-library (10x1) attached.

- Can backups of the lvm snapshot be made through SAMBA/SMB or NFS ?
-> better with Linux server pushing data via rsync(?) to a SMB or NFS share on the Windows machine and then back it up (disk-to-disk-to-tape strategy)
or
-> Windows machine pulling data from a SAMBA or NFS share on the scalix server (that points to the snapshot) within the backup process (disk-to-tape)

The former requiring much more space but should be faster, the latter viceversa... but are SMB or NFS viable in any case?

- Any other suggestions/directions on this (or other) topic would be greatly appreciated... !!! :)


Please be as much verbose as you can because this is the first "big" installation we make.

Thank you and keep up the good job


--
Davide DG.

I'm not able to answer all your questions, but I can share my experience about the performance.

I'm running Scalix (currently 10.0.0.5 so no caching) on a dual Xeon 2Ghz machine with 4 GB RAM.
Unfortunately my budget is not so big, so I decided to use 4 SATA disks in a (hardware) RAID 1+0. I have 150 premium users connecting with Outlook 2000, about 50 standard users connecting throug IMAP (Thunderbird) and 20 standard users connecting through SWA. We have a number of users with very large mailboxes (up to 12 GB..). Total mailstore is around 160GB at the moment.

The performance of this machine is good, in fact better than I have hoped for.

[adhodgson]

Hi,

We are going to be using a Raid-10 1TGB system, so what we will do is to LVM 500GB for the Scalix and use the rest for snapshots. We will suspend the Scalix server, then take the snapshot, and then Tar the snapshot and rsync this to a Windows server, which will be backed up.

You can't just rsync to a Windows server due to permissions issues.

We have around 150 users currently.
Thanks.
Andrew.

[davidedg]

You can't just rsync to a Windows server due to permissions issues.

and maybe with NFS (not SMB) Server services on the Windows machine (it's a 2003-R2 system)? Anybody tried this one ?


Or, viceversa, the Windows Server acting as an NFS client and grabbing the data from the snapshot (which should be NFS-exported) ? But, this way, I should consider using a script and/or a timeout to notify the linux server that it can dismiss the snapshot once the backup has finished or any error occured... uneasy I think.

Thanks.
--
Davide DG.

[kanderson]

Your server hardware will be fine.

SCSI drives are very fast. In spite of the comparisons often made between SCSI and faster SATA and/or IDE, SCSI will slaughter them during periods of sustained high use, as well as reliability. SATA is gaining ground, particularly with NCQ, but SCSI is still much faster under load. I wouldn't be worried about it, to be honest. Especially if you have a good controller with some cache on it. Remember, mirroring is exceedingly fast for reading, and that will be the majority of your work.

Scalix Enterprise Edition connects two (or more) sites very easily. see here:
http://www.scalix.com/wiki/index.php?ti ... alixServer

If you can, I'd suggest a set of small mirrored drives for the OS, and use the 160s alone, mounted as /var/opt/scalix. I'd be more worried about splitting it out like that than splitting out temp ot tmp for Scalix. But that's just me, and I've already said I don't think you'll have a problem. with disk speed.

The recommendation for free disk space is 25%. In my view, this is like 2x RAM for swap or 5% reserved disk space on ext2/3. It's a good baseline a few years ago, but that's alot of gigs of disk. Far more than I could see needing. If you want to only assign 5 to 10 Gigs, that would still be TONS in my opinion. I guess with 25, you could grow the FS if space gets tight, and buy yourself some time to get new HDDs.

I'd look at this doc for LVM info:
http://portal.knowledgebase.net/utility ... ?rid=26548

The BEST solution is to Tar it up on the Linux box, so permissions are saved, and then copy it over with Samba or Rsync. Or FTP if you can automate it. That will allow the fastest recovery. Having said that, you could certainly recover from a backup which loses permission data by running "omcheck -s -d > /tmp/chkscript && bash /tmp/tmpscript" after you restore and reinstall Scalix. That script just takes a long time to be created and also a very long time to run. But it'll get you going...

Push, Pull, whatever you like best. It doesn't really matter.

NFS might work too, but a single file will copy faster than several small ones, which is why I like the TAR option. Tar will also perserve your permissions which will speed recovery as mentioned above.

Kev.

[davidedg]

Thanks for the precious info.

I think I'll resort to tarring the lvm snapshot over the network to a SMB/CIFS store (the windows backup media server); then have the resulting tar file backed-up to tape by Veritas BackupExec.

About the RESTORE process... with *only* this full backup, can I restore it (or part of it) on a "restore" mailnode and then export the messages/folders I want to restore? and how?

I mean... having only a full tar backup, can I restore a single mailbox either directly to the "current" mailnode (for example for full mailbox restore after corruption) or to another mailnode/mailbox for retrieving single messages ?
Does this envolve any special step? Or... should this be done in another way?

Thank you for support.

--
Davide DG.

[kanderson]

Single User restore instructions can be found here:

http://portal.knowledgebase.net/display ... =0.7947962

However, I find it much easier to use sxmboxexp to dump all users (omshowu -m all), so tthat a dump file is created for each user. From there, as you indicated, you can restore the dump file into a user "Restore Restore" or someone in another mailnode. Then sign into that Restore user's profile, and pull out message as neccessary.

In addition to sxmboxexp (and sxmboximp for importing), Scalix 11 also supports recovery folders. Recovery folders work similar to a recycle bin, where a deleted messages is hidden from the user, but still available in a folder called recovery. This folder can be made viewable by the admin through the Management Console. By default, it will keep deleted email for 7 days. If you have the disk space, that number can be jacked up to 30 or more. This is a hugely beneficial feature, and greatly simplifies life for admins. Having said that, it's only available for Premium licensed users. So even Community's premium users don't have this functionality. Depending on your circumstance, this may be beneficial for you. I consider it the #1 understated feature of sx11, which in my view contains several compelling functional increases. I hear about smartcache regularly, but I never hear about this. But it's huge. So are sxmboximp/exp as compared to omcpoutu/inu, which no keep rules, and user settings, but also allow the export of public folders.

Kev.