Data Domain 2X00 vs NetApp E-Series

[ninjaburn81]

So looking for some feedback...

We currently are running a Data Domain DD620 and are basically out of space. I've done everything I can but it's time to upgrade, and it is a shame because the unit is rock solid. So we are currently evaluating two options:
1) EMC Data Domain 2200 (or possibly a 2500 but I don't forecast our data growth to be enough to warranty the extra cost of the possibility of adding additional trays)
2) NetApp E2700 series

I must say I love the Data Domain boxes and ours has been rock solid, but the NetApp integration is intriguing. Still awaiting back quotes on the NetApp but I imagine they will be in the same ball park so all costs being equal, what does everyone think of their DD and NetApp boxes?

A few notes:
1) we are looking at the E-series, not the FAS so keep that in mind.
2) This will only be as a backup target
3) SMB, so I want hands off as much as possible in regards to maintenance.

Thanks!

[dellock6]

Adam, I'm no suggesting one is better than the other, but remember the integration we have in the software is with NetApp FAS arrays, not the E-Series, and even if it would be supported, you are talking here about backup repositories not the primary storage. On the other side, we tested the E-Series as backup repositories and they are really good, performances are really high (also because it's not deduplicated as the DataDomain).

[ninjaburn81]

Luca:

Thanks for pointing out that the E-series lacks the direct integration, that was not shared with me. I believe Veeam has a similar integration now with Data Domain now, does it not?

Performance-wise we are doing very well now with the Data Domain box so any speed boost is just gravy; my main concern is purely storage space. That is one reason I am leaning toward Data Domain is the dedup helps tremendously in that area and NetApp's offering would lack that.

[foggy]

Seems you're confusing two types of storage integration: backup from storage snapshots (that we already have for NetApp and will have for EMC's VNX/VNXe storages in v9) and integration with deduplicating storage system as a backup target (that we have for Data Domain).

[ninjaburn81]

Thanks foggy for the clarification, and yes I must be confusing the two.

[Gostev]

Ideal solution here would be E-Series as primary backup storage, and Data Domain as secondary (for long term retention).

[ninjaburn81]

Ideal solution here would be E-Series as primary backup storage, and Data Domain as secondary (for long term retention).

Ideal = $$$

I am looking at just going at the DD2200 in the end, along with the DD Boost.

So here is a question...and I can open a new thread on this...does DDBoost help at all when using Veeam Endpoint pointing to a B&R repository sitting on a DD box??

[Gostev]

Sure, it does help.

[localhost]

Just as you know DD2200 works beautifully in backup, it is terrible in restore (i.e. HOURS to even browse the file in a backup, 1 DAY to browse using Veeam explorer for Exchange. This is BROWSING not even RESTORING). I hihgly recommend to run a POC with the option to the return the DD before going into production.

[smpartridge]

I have a pair of DD2500s that I use as backup repositories. The backup speed is decent, but the restore speeds are very painful. I would suggest that you look at getting an x86 server and load it up with local drives. Then it can be your Veeam server and local backup target. The speed will be far better then the DataDomain and much less expensive. I have done this and now only use the DD2500s as a target for backup copy jobs for long term retention. They work well for that, but it is still a slow process to restore from them. Just my two cents, but it works great for me.

[Dave1337]

We have a pair of replicating DD2200's. The backup performance is great, restore performance is terrible. Almost scary slow.

Maybe someone can help?

Configuration:

DD2200's using DD-Boost - Code level 5.5.0.8-456696
Veeam - Code level 8.0.0.2030

Enable inline data dedup - disabled
Exclude swap file blacks - enabled
Compression level - Optimal
Storage optimtization - Local target

I believe those are all the settings related to the backup repo.... This is how we were told to configure it when we implemented. Restore performance is terribly slow with this config. Any tweaks I can make to increase performance? Max restore performance is like 20-30MB/s, sometimes even lower than that...

Thanks in advance and let me know if you need any additional info.

[emachabert]

Some tweaks regarding the backup configuration:
Compression should not be turned on ( or in dedupe friendly mode only). This setting won't affect performance much but the dedupe ratio for sure.
For the restore speed, it depends if you are speaking about item restore or full vm restore. In the first case, it is not so bad for a little 2200.

[smpartridge]

Agreed that the backup performance that I was getting with the DD2500 was decent, but the restores.... Scary is a good way to describe it. If you want to see something really funny, or not if you actually need it, is to do an instant recovery from the DataDomain. It is not a pretty site. Again I believe that that DataDomains are great at dedupe and for long term archiving, but not as a primary tier for Veeam.

[jveerd1]

Restore performance is one of the reasons I would recommend to look at other options like ExaGrid. With their landing zone architecture they might be able to solve your restore performance problem.

You really want to utilise a product with Veeam integration, be it DD Boost, Veeam-ExaGrid Accelerated Data Mover or a standard Windows/Linux box. Please stay away from SMB as a backup target if you can.

[emachabert]

Deduplicating appliances aren't designed to handle random read workloads, therefore aren't good at everything that will trigger that kind of workload (eg. INstant recovery ).
The best is to have a landing zone using traditionnal block ( DAS or SAN) holding the short term retention (15 days to 1 month for example) and then having a third level for mid term retention (1 week to 1 year for example) using deduplicating appliance through a backup copy job, with Veeam integration(Dd boost for EMC or catalyst for HP in v9).

The problem is that the best isn't the cheaper, and often the money take the lead and you prefer investing in the mid term retention.

[dellock6]

However the first tier to ingest random IO (happening also during incremental backups) doesn't need to be large, it just have to hold around 1-2 weeks of backups depending on the design. Thus it can be a fast and costly storage, since the third parameter (space) will be low. Then, you can couple it with a dedupe appliance, where speed is the bad parameter (compensated by the fisrt tier) but gives you cheap space.

[mongie]

Ideal solution here would be E-Series as primary backup storage, and Data Domain as secondary (for long term retention).

Not sure how much you guys deal with EMC on a solutions level, but we're currently being sold a new storage solution for our backups, and they consider the DataDomain as the only storage you need... We're looking at the DD2500.

It doesn't sit right with us based on previous experience with dedupe appliances and veeam, but thats their solution...

[dellock6]

Alex,
we are aware EMC suggests to use the DD as the only backup target, and from their point of view, this is totally understandable.
As other users have said, depends on what restore performances you are expecting, for the ingestion part there are no specific issues.

[mongie]

Restore performance would only be better with faster tier if you're restoring from recent backups though - right?

[dellock6]

Any data stored in a dedupe appliance needs to be re-hydrated during read operations, regardless their age. It's why me and Anton and others were suggesting afirst tier made with a non-deduped system would be better, if the concern is around restore times. If they primary goal is reduction of consumed space of backups, then a dedupe appliance is the best choice.

[jveerd1]

Except for the ExaGrid dedupe appliances which use a landing zone architecture to store your most recent restore point(s). This way your youngest restore point(s) does not need to re-hydrate, thus faster restores.

The landing zone architecture might make the ExaGrid a good first tier choice. Depends on how much "older" restores your business needs to perform and what kind of SLA is in place for restores.

By the way, this comment is just to complete the picture. I have no interest in ExaGrid whatsoever.

[dellock6]

Correct. I've left ExaGrid out of my reply as it was not listed from OP among his options...The dual design I've suggested is somewhat resembled in the ExaGrid architecture.

[emachabert]

Having tested Exagrid I am mitigated with the Exagrid landing zone in the case of Veeam because it won't be useful in many restore use cases.
The landing zone is holding the most recent backup points and is speeding up the write process since there is no inline dedupe when writing to the array.

But when it comes to restore, you rarely just hit the last restore point because you are in an incremental chain and the full won't be in the landing zone. If you are doing a simple file restore targeting a recent backup point, you will be ok. But if you are using any Veeam Explorer (Exchange, SQL..) you will trigger the rehydratation of data from the full and other deltas and it will be as slow as the competition.

I always try to have a landing zone using no deduped volume on DAS/SAN with a good performance (write cache/good number of spindles) to permit the use of all Veeam Advanced technologies (Instant recovery/vPower NFS/SureBackup/Explorer etc.). This landing zone is sized to allow 15 days to 1 month of daily backup retention. Then I use a deduplication Appliance to store the mid term and allow bandwidth optimized replication to DRP site.

In v8 the best was to use a backup copy job to a DD Boost enabled array because of the optimized copy (no blocks being moved, only pointers). I am very happy that Catalyst will be supported in V9 because I have always had better restore throughput with the Storeonce than with the DD.

[ericcsinger]

Me, I would not go with eaither solution. If you checkout my blog http://ericcsinger.com I'm actually slowly writing a backup storage series.

I would personally reccomend Microsoft storage spaces. My current solution can do up to 6k iops, 3gbps write and 5gbps read (those are bytes not bits) per pool. I have a tad over 100tb usable per 80 disk storage pool and I have two storage pools. Total solution probably cost under 75k (have to go back and look at quotes).

That said if you want tier 1 storage, I'd choose nimble storage over netapp any day of the week. The only value add to dedupe targets is when you keep over a month of retention on disk. Otherwise general disk is a better value.

[emachabert]

75K looks a lot. With that you can have a very well equiped MSA2040 with dual controller, 16Gb SAN connection and so on.
I have installed one with 146*1TB SFF 7,2K and it costed 50K, and it has very impressive perofrmance for veeam in reverse incremental (configured with two wide stripped volume, one per controller).

[dellock6]

Looks a lot to me too... However, the comparison with a DIY (do it yourself) solution and a buy solution is not always fair, you have to count the time spent to research, design, order components, and build the final solution. Man hours are not free, and sometimes require skills that not anyone has. This is something separated from the dedupe yes/no debate.

If you then wants to discuss a server-beast kind of design, that's another story, and I can show you easily how you can design a machine with almost 200 TB usable space, a 2TB NVMe card in front for caching, for less than 30k.

[emachabert]

I have some of these setups, including ones with the smartcache also
there are many ways to build the repos

[ericcsinger]

Looks a lot to me too... However, the comparison with a DIY (do it yourself) solution and a buy solution is not always fair, you have to count the time spent to research, design, order components, and build the final solution. Man hours are not free, and sometimes require skills that not anyone has. This is something separated from the dedupe yes/no debate.

If you then wants to discuss a server-beast kind of design, that's another story, and I can show you easily how you can design a machine with almost 200 TB usable space, a 2TB NVMe card in front for caching, for less than 30k.

That's less than 75k, its not white box junky HW, I have full redundancy from top to bottom.

4 JBOD's 60 disk bay in each, all JBODs have dual controllers.
160 disks (4TB) each
2 full blow servers (that double as proxies). The servers each have 2 quad port SAS cards, that have direct connections to the JBOD's (SAS MPIO) no daisy chaining. They also have dual 10g networking in an LACP config.
The Windows Storage pool is clustered.

I can lose an entire enclosure, a controller in the remaining JBODs, a server, a NIC and the system would still be running and accepting traffic.

This HW all has 5+ years HW warranty and support (Raid Inc. is the JBOD and disk retailer) and Dell is the server.

As for complexity / man hours. That's a fair statement. This was more of a "you don't need to run tier 1, but you also don't need junky HW, there is a middle ground" sort of comment. I'm going to be documenting everything I did anyway, so someone will easily be able to duplicate it.

Also to reply to emachabert, 160>146 and I have 4 times the raw capacity per disk and LFF 7.2k perform better than SFF 7.2k. My usable capacity is after a RAID 10, and leaving a recommended 30% free space in the pool. The storage is massive, fast, customizable, resilient, supportable and affordable.

[emachabert]

Nice setup.

Don't want to enter any argument here, but you can't say a 4tb lff drive performs better than a 1tb sff drive. Generally speaking, seek time and latency are lower on sff and the smaller the capacity is, the lower the seek time also is. Mtbf is higher on lff.


I stay tuned to read your blog post

[ninjaburn81]

We are now considering a DIY (or at least a maxed out storage server) to use. We have requirements that have us doing LOOOOONG term retention, so I am using tape for that. I have been doing the math on our growth curve, and unfortunately I will be needing the expandability of the DD2500 to manage it all in the next few years. (The 2200 cannot add trays.)

Anyone know of threads that talk about a DIY or storage servers people have had vendors build? Love to get some reading done on it. I know Gostev has had some notes in Community Digest, so I am looking there too. The big thing for me is de-dup is a wonderful bonus for us, since my users are notorious for keeping multiples of the same file EVERYWHERE.