Suggestions wanted

[localzuk]

Right, I'm looking at my server replacement plan for the next couple of years (as some of the servers are now hitting 3 years and therefore are falling out of warranty).

 

My ultimate plan is to move to a virtualised infrastructure with a SAN for the storage - the only issue is that our backup solution at present is a server with giant hard disks in it, running Backup Exec.

 

What would be a good backup solution to replace that, but with the future capability to also backup our virtualised system? I'm thinking a SAN for this would be a bit overkill. A NAS device, with a cheap server running the backup software (whatever we end up using).

 

Another thing to note is that we can't afford something super-fancy like the Sun S7000.

 

(So you have an idea of size, we are a middle school, total backup size for data is currently 300GB (ie. that's a full backup, so we'd want a few months worth storing, plus weekly backups, plus daily incremental)

[Duke]

I'm in a very similar position and currently planning to do the virtualisation next summer then migrate to a freshly-built virtual domain and Win7 the summer after. However, we're a fairly large IT Specialist school so do have a bit more budget to play with than most. A full backup for us is 1.2TB, although that'll grow once I can bare-metal VM snapshots, then shrink once I can de-dupe them.

 

My plans (I know these don't be completely relevant to you but might give you something to think about) are currently like this:

 

Virtual Machines are hosted on a 7410 with clustered heads and the are replicated to a 7310 elsewhere on site. The 7410 will handle snapshots of the virtual machines. This gives me redundancy in that I can have no downtime if a 7410 head fails, and if an array or both heads go down I'll have failover to the 7310.

 

I'll also be using Backup Exec on a backup server connected to a NetApp FAS2020 to handle both virtual machine backups and granular backups of the actual data (e.g. using the Exchange and SQL BE Agents) in case I need to do a file-level restore rather than a whole server restore.

 

Somehow tape will also come into this, whether it's an autoloader hung of the back of a 7410 via NDMP or whether Backup Exec handles it I've yet to decide.

 

On a budget... I'd check out Bacula and Amanda and see if they might fit your needs. I've never used either personally but have heard good things. The advantage of staying with Backup Exec is that (for a high price) you can get the VMware Agents.

 

What SAN were you planning to use for the virtualisation storage? If you can't get an S7000, any idea if what you do buy will support snapshots?

 

A cheap NAS and server for running the backup software sounds like a good idea, just make sure you have some level of redundancy in there for disaster recovery. If your server room burns down will it take the VM SAN and the backup NAS with it? If so you're stuck, and need to find a solution like a tape drive so you can put the tapes elsewhere, or an external HDD you can archive the backups to then take out of the server room.

 

Chris

[tmcd35]

I use a 4Tb NAS box onsite and do regular backups to that. Each VM is running backup software (Yosimite in this case) and does file level backups daily/weekly/etc.

 

I then have a seperate 4Tb 'portable' NAS that I keep offsite. During the summer/xmas/easter hols I down the VM's one at a time and copy thier .vhd files to this NAS box manually.

[DMcCoy]

You need to be careful with snapshots. If you are using the SAN to create snapshots then it should be less of an issue, but with VMware etc it will create an additional disk files for the subsequent data, this can become very large with all data written since the snapshot in it.

 

Should the redo file become large enough it will no longer quickly or possibly successfully be merged back into the main VM file. I once left a smoothwall VM in snapshot mode by accident after an upgrade and it was painful getting it to merge back the changes. There can be some not insignificant additional overhead when running with a redo log too.

[plexer]

Look at the R1Soft CDP for your backup does virtual full backups, ver 3 thats coming out doesn't need a dedicated server to be the backup server as you have with the current ver 2.

 

Ben

[richardp]

Hi check out the DroboPro which is around £850 with no drives and has a single iSCSI connection or if you can find it the new DroboPro Elite which has two iSCSI connections. Both have been officially tested to work with VMWARE. So for your current needs it at the very least provide you with a cheaper backup for the SAN (just add upto 8 sata drives to suite your requirements) or even be suitable to be your main SAN.

 

richard

[plexer]

Hi check out the DroboPro which is around £850 with no drives and has a single iSCSI connection or if you can find it the new DroboPro Elite which has two iSCSI connections.

 

1 iSCSI and 2 iSCSI ??? iSCSI is not a physical connection?

 

Ben

[Guest Guest]

Ive been looking at this problem lately. Ive come to the conclusion that there is no getting around using block level deduplication.

 

R1Soft does it. And so does the latest version of the ZFS filesystem. Opensolaris will have it soon, solaris already does.

[IrritableTech]

Another thing to note is that we can't afford something super-fancy like the Sun S7000.

Have you actually had a quote?? I'm told Sun education prices are much lower than list prices.

[Duke]

1 iSCSI and 2 iSCSI ??? iSCSI is not a physical connection?

 

Could mean two NIC ports dedicated to iSCSI maybe? Or maybe just 2-port Gig NIC. Know what you mean though.

 

Have you actually had a quote?? I'm told Sun education prices are much lower than list prices.

 

Agreed, Edu Promo will save you loads, as will matching Grant if you need more than one. Give Andy at Cutter (linescanner) or Andy at WTL (Apaton) a shout.

 

Chris

[localzuk]

Have you actually had a quote?? I'm told Sun education prices are much lower than list prices.

 

Looking at the matching grant program, 2 7110's would cost us £8600. For 1.7TB in each unit. That is a lot of money!

[richardp]

1 iSCSI and 2 iSCSI ??? iSCSI is not a physical connection?

 

Ben

 

Sorry obviously did not make myself clear, the drobopro has a single gigabit network port, the drobopro elite has two gigabit network ports. Add one or two gigabit network cards to your server approx £80 for an intel single port card, and you can connect the drobopro directly to the back of the server using a standard network cable, or to your network switch. This means you can site the drobopro next to your server or pretty much anywhere else on your LAN within usual restrictions on cat5 cable runs.

 

You then install the provided iSCSI initiator software on the server which allows the server to communicate with the drobopro using the iSCSI protocol. You can then create volumes automatically or manually as you require.

 

Richard

[Hebdenlad]

can yo give me a call? I think I've sent you a PM but not sure.... long day and toop much technology :-)

 

Cheers,

 

Phil

 

01252 420 000 and ask for Phil Lawrence

[diggory]

I'd second the drobo option.

 

Running a drobopro as an SAN off an ESXi box.

It isn't the fastest thing around, but very flexible in it's ability to take off-the shelf sata drives in a mix and match format with hot swapping

 

On a few draw backs...

 

The drobopro can only be configured with the drobo control panel software - which needs to be installed on a mac/linux/windows box.

It's a bit slow.

 

I'm using it to run MS DPM on it and ESXi box, to get around MS not supporting iSCSI and dynamic volumes under server 2003.

Also watch out the although drobo support ESX, they don't support dynamic windows volumes - only basic (!)

[diggory]

I'd second the drobo option.

 

Actually, our drobo-pro is having problems running for long periods of time, so I'm not so happy with it (as well as being slow to run one vm).

 

It seems to drop the connection after a random amount of time (between a week to several weeks).

Ho hum...

Edited June 21, 2010 by diggory

[dhicks]

Ive been looking at this problem lately. Ive come to the conclusion that there is no getting around using block level deduplication.

 

I'm replying to an old post here so, with the benifit of hindsight, I'm going to be avoiding block-level deduplication for the moment. I tried OpenSolaris on standard Intel hardware and got data corruption problems with disk images being stored on ZFS deduplicated volumes - it might well be that I'd done something wrong, but I couldn't afford to spend any more time figuring out whate the matter was. I then switched back to Linux, which was reliable, however the de-duplicating file systems available for Linux are more on the "experimental" side and run as FUSE modules, which gave dreadful performance.

 

In the end, I figured the easiest thing was going to be to write my own file-level deduplication script on top of a standard EXT4 filesystem. Nice and simple. Hardware-wise (just going back to the original post), I plan to install a large Antec 1200 case stuffed with 12 2TB harddisks over the summer, running Linux, rsync and my own file-level deduplicating Python script as a backup solution. 22TB of backup storage should hopefully last us for a while.

 

--

David Hicks

[diggory]

Might be worth trying nexenta?

It's open solaris with a gui though... so might not do it?

[Hebdenlad]

Hmmm. The S7000 does de-duplication as standard (it uses ZFS and Open Solaris with an intuitive interface). It does block level de-dupe and you can turn it off and on at either a share or project (read template) level.

 

The advantage of investing in a product such as the S7000 from Oracle is that it is fully supported, installs and works in a very short space of time and has an excellent roadmap going forward (am I sounding slightly biased here?).

 

Ask the other edugeekers about it and see what response you get.

 

Don't forget that you'll lose some of your 24TB capacity to RAID and the S7000 has, because it uses ZFS for RAID, the best usable capacity of any comparative RAID system.

 

If you need / want to chat about it then please drop me a line on here or reach me via the Oracle main switchboard.

 

Cheers,

 

Phil

[fox1977]

How much are we looking at for a S7000 Hebdenlad? Rough price is fine.

 

Just looking at options on a few boxes and the S7000 comes up quite a bit

[Duke]

See your other thread, Andy should be able to get you an accurate quote based on your requirements.

 

EDIT: The other thread is HERE for those reading this.

Edited December 3, 2010 by Duke

[Hebdenlad]

As you can see Duke has pipped me to the post on a reply. He must spend all day just watching the threads on here. In terms of a price it depends on how much capacity / performance you need / want.

 

The S7120 starts at 12TB using 1TB drives (there is a 2TB driver version as well) and can scale out to 60 drives. Moving up the range we can support clustering for high availability, read and write optimised SSDs for massive IOPS and with the top model you will be able to scale out capacity wise to just under 3PB.

 

Andy on here should be able to give you a good indication of price / capacity / performance matched to your budget. I'm the technical one for the S7000 range.

 

If I can help any more then please let me know and if you need me to come and see you I'm not that far away from Liverpool - live near Halifax and am always glad of local work.

 

Regards,

 

Phil

[multipoint]

Microsoft and VeryPC launched their EcoVault range which is pretty well priced? VeryPC, Green PC | Broadleaf | Greenhive | Eco PC

[zippo]

Just a word of warning regarding the use of snapshots as a method of backing up Domain Controllers. Please remember that it is possible to really screw the Active Directory after restoring a snapshotted server. If the AD time stamps are out of sync for to long - the system will mark the server as having a degraded AD and that will effectively stop AD replication and updates. Whist not fatal - it can be a right pig to fix and involves the forced grabbing of FSMO roles and the forced removal of the server from the AD. Before adding it back in and re-allocating ther various roles.