Need architecture ideas with dedupe and tape in mind

[dgapinski]

Hello,

I am looking to purchase a new dedupe appliance (HP StoreOnce) and at my org, tape copy is also a necessity. From my research and from talking to one other admin, it seems that in order for copying to tape will work in a timely fashion, one must:

1) Run a backup job to the dedupe appliance
2) Run a copy job to copy the backup from the dedupe appliance to a plain old LUN
3) Run a tape copy job to copy the backup from the LUN to the tape library

With a much older DataDomain, I have tried skipping step 2 and copying from the appliance straight to tape, which would take more than a handful of days to do. But I have to ask, is this 3-step method still necessary because of poor rehydration performance from the dedupe appliance to tape library? If so, place me squarely on Team Exagrid.

Many thanks for your thoughts!

[tommyo]

Take a look at the DXi appliances by Quantum. They have some ability to go Path to Tape and they also have the Veeam Datamover Service installed on the appliance itself which greatly improved performance for us. They also just came out with Scalar i3 which you can purchase with a blade server in it that runs Veeam.

[gurneetech]

I've been looking at what I need to do for a 'next generation' Veeam backup environment as well. We initially deployed Veeam in 2010, and as much as components have changed, its largely the same operationally as it was 7 years ago. Both the StoreOnce and the DXi are on my radar, my guess is that pricing will drive the actual selection as they both seem like good options.

One thing I did see looking at the HP StoreOnce, is that it has a pretty low backup-chain limit. My back-of-the-napkin concept for the HP product looked more like this:

1) Run a backup job to backup server with DAS
2) Run a copy job to copy the backup from the DAS to the dedupe appliance for longer term storage (daily)
3) Run a tape copy job to copy the backup from the DAS to the tape library (weekly/Monthly)

I haven't digested the implications of the DXi integrated datamover service yet, but I don't think it has the same super backup chain limit.

[dgapinski]

So regular array storage was required for you as well. can you tell me why or what you observed to configure it that way? Any stats you could offer would be most helpful!

[ChrisSnell]

Dan, seems you're already familiar with ExaGrid. The landing zone architecture will allow you to keep a weekly full on 'normal disk' for a week, so that you can push to tape during that time. Because you don't need to rehydrate, the performance is very fast - and features the Veeam Data Mover too. This is a very common architecture in use with health care organisations.

Obviously there are many other benefits from using the ExaGrid appliances - Instant Restore speed being one, along with scale out architecture (like Veeam), SOBR support and others.

Feel free to PM me with any questions.

[dgapinski]

I am familiar with Exagrid - the platform looks like it's more expensive to expand, but far simpler to configure and a better performer all around. My company is rather entrenched in HP, so I'm interested in which piper I have to pay if I go with a StoreOnce. It seems to me that if you dump to tape, you need "a" landing zone no matter what you use. So there's extra steps (by having to introduce an extra copy job into the schedule) with tape offload and tape restoration. I'm trying to see if anyone in HP land has had a different experience with their units though. Doesn't look like it. Prove me wrong?

[skrause]

From what the HP engineers told me when we were looking at the StorOnce several years ago is that their deduplication can be enabled on a per-volume level. So you would not necessarily need to have a separate storage array just for your non-deduped copies.

You may also want to see how well the StorOnce performs just pulling the data off and going straight to tape rather than copying it to another location.

[dgapinski]

It's worth a try if I have to get that, but I'm really looking for field experience so that I can avoid a purchase that is more of a hobby-creator. Anyone else who has worked with it? Is the water ok - with pulling from a store-once directly to tape?

[ChrisSnell]

I am familiar with Exagrid - the platform looks like it's more expensive to expand, but far simpler to configure and a better performer all around.

I think you'll be pleasantly surprised when you look at the costs. The major cost in the inline scale-up solutions is the compute. They need a lot of grunt to dedupe the backup stream, so are always using the latest processors, and lots of them. They even pass some of the dedupe off to the clients being backed up(!).

ExaGrid does require more storage, but that has become commoditized. Less CPU is needed because the dedupe is not inline. ExaGrid also features a price promise based on the original sale price - so you can predict future costs for 5 years. Take a look at this hospital case study: http://www.exagrid.com/wp-content/uploa ... -Story.pdf

[skrause]

It's worth a try if I have to get that, but I'm really looking for field experience so that I can avoid a purchase that is more of a hobby-creator. Anyone else who has worked with it? Is the water ok - with pulling from a store-once directly to tape?

Have you contacted your HP reseller about doing a POC?

[dgapinski]

I have actually. They can offer a virtual appliance to test with. Normally that would be no problem, but I am dealing with a little static on the supervisory side - when I mentioned testing with the virtual appliance, I was denied the opportunity to do so, which is why I am asking about field experience, as a last resort for info gathering.

[Delo123]

Why not use a full-blown Windows Server with Deduplication. HGST has a new JBOD 1224TB in 4U under 50K. Veeam will probably store up to 10PB on that single box at an unbeatable price...

[dgapinski]

Ahh the voice of reason! I actually was thinking along those terms after I wrote the last response. Since we have been using ReFS dedupe backups on a smaller scale, I ran a test backup to tape over the weekend and found that the backup times are comparable, so I'm thinking that might be the better way to go. Good thought Delo!

[reed76]

You have done exactly what we had to do at my last gig; backup directly to dedupe (EMC DD and Dell DR4x00) then pull from there to tape. It was slowwww (even the newer faster Dell).... We really needed to have a disk LZ from which to do backup copies to dedupe and tape at speed but management was thoroughly sold on direct to dedupe (thanks, sales guy) and it stopped there. The company itself was sold and we all left before I could eventually convince them to get another cheap SAN for the LZ. While I never worked w/ one, people always claimed the Exagrid was made exactly for this and worked at it quite well.

[skrause]

Why not use a full-blown Windows Server with Deduplication. HGST has a new JBOD 1224TB in 4U under 50K. Veeam will probably store up to 10PB on that single box at an unbeatable price...

Windows server dedup is actually pretty good for a tape source since it is post-process and you can set how old things are before they are processed by the policy (default is 3 days).

It does take some thought about how many/how large your volumes are that are being de-duplicated since Windows dedupe is single threaded per volume. It will run multiple volumes in parallel so more volumes is sometimes better, though you don't get the same overall dedupe ratios.

You also will likely want to architect your jobs in a way that does not do synthetic operations (periodic active fulls), especially on large backup jobs.