Netbackup Appliance vs DataDomain

[Mastana420]

Hi

We are in a process to make changes to our current backup environment. Our main focus is to do some changes in our 2 main sites. Both sites have their respective domains with multiple mediaservers (no appliance or DD). What we are trying to achieve is as follows:

1. Try to merge both domains, so both are member of the same domain acting as a cluster with 1 node at each site.
2. Purchase storage or appliance to achieve the following:
   With 1 storage\appliance at each site, have optimized duplication enabled so we have 2 copies of the backup, one at each site (storage\appliance) at all time. This to gradually go away from weekly tape vault and go over to a disk based solution.

I have done som research already, and this forum thread is to verify the information in hand and to add even more information is possible.

Could someone give me answers to the following questions:

• First and foremost, to achieve the above mentioned goals, Do you guys suggest DataDomain or Netbackup Appliance? And then why?
• Is optimized duplication only a feature available in storages like Netbackup appliances\DD or is it something you can achieve with two MSDP pools at each site?
• Netbackup recommends that we segment the data on a dedupe pool for best performance, how are we suppose to achieve that when we only can create 1 pool out of 1 Netbackup appliance? In DD you can create mutiple pools, and have a Global deduplication enabled. How can we achieve that with Netbackup appliance?

[GulzarShaikhAUS]

Merging two domains is possible through catalog manipulation which can be carried out by certified partners. But it wont be possible to have two active master server on two sites with common catalog or I would say it will be challening and practically impossible.

A better approach to your scenario would be two different domains running with Netbackup appliances with cross site replication with AIR. Thus at any time you have data protected at each from both domains.

[GulzarShaikhAUS]

About achiving global dedupe using Netbackup appliance, it wont be neccessary unless the required capacity exceeds maximum capacity Netbackup appliance can provide. 

The newest Netbackup Appliance model 5330 provides upto 228TB with single appliance with disk shelves. A bit cheaper 5230 can support upto 148TB.

[RiaanBadenhorst · Accepted solution]

Hello,

Lets review your requirement

1. Try to merge both domains, so both are member of the same domain acting as a cluster with 1 node at each site. This is possible but only one node will be active at any one time.
2. Purchase storage or appliance to achieve the following:
   With 1 storage\appliance at each site, have optimized duplication enabled so we have 2 copies of the backup, one at each site (storage\appliance) at all time. This to gradually go away from weekly tape vault and go over to a disk based solution. This is possible.

For point 1 above you have to understand that for this scenario your link between the two sites is extremely important as it connect the media servers and clients to the master. If you lose the link you'll not have backups/restores in the site where the master is passive. Just something to keep in mind.

Depending on the type of clustering you're referring to their are more considerations. The alternative to this is using AIR as mentioned above. With AIR you'll have to separate catalogs with the ability to selectively replicate (duplicate) the backup images between them. This allows you to restore images from one domain in the other if need be.

Please note that the AIR configuration does not protect you against any failures of the master, media or storage devices in one domain. If these service in one domain go down, the clients are not automatically protected by the other domain (its possible to reconfigure them but can be a bit of a task). So its good to have a level of high availability / redundancy in each doman to ensure backups continue.

Lately the DD/Appliance comparison is basically equal, almost all features exist for both. The only one that could be an issue is SAN Client. If you're going DD you''d  need Linux media servers to be installed (if you're not running linux already).

HTH

[Mastana420]

Hi. First of all thanks for the feedback. It is highly appreciated.

I have had a Symantec technician analysing my current backup environment and analysing how to utilize site2 better. Regardless of merging the two domains or not, i need to achieve some kind of cluster because our current environment is very vulnarable with only 1 node on our main Master Server. I can achieve clustring by following methods:

• By putting one more node in Site1, with 1 active and one wofking as passive and taking over in case a failure occur
   
• Or by merging the 2 domains (site1 and site 2) which will result in 2 nodes, where the Master server in site 1 will be Active, and Master server in site 2 passive, and taking over incase of a failure.

What the Symantec technician suggested was to merge the two domains. We have a very good link between the sites so that should not be an issue. As for your suggestion about utilizing AIR, it is a good alternative, but how about the performance? Will i be able to achieve optimized duplication by utilizing AIR? Optimized duplication is a key feature for us to minimize the bandwidth usage as much as possible, and achieving duplication on the other sites realy fast.

[RiaanBadenhorst]

Optimized duplication is used by AIR. You duplicate within a domain, you replicate between domains. It uses the same underlying process to do both. Also note that for the one domain option you'll need Symantec PS to merge the two domains which would be an additional cost to think about. Not a bad thing, just keep it in mind.

[Mastana420]

Cool. Another big setback utilizing AIR in two domains, is the following:

Although Auto Image Replication is a disaster recovery solution, the administrator cannot directly restore to clients in the primary (or originating) domain from the target master domain.

I assume this is not the case with the other solution, by clustring between two sites.

[GulzarShaikhAUS · Accepted solution]

I think this is not the case as you can very well restore from other domain. You just need to establish connection between the master/media server and the client and change the Master server entry in the BAR GUI. It shows the copies from other domain and then you can perform restore. I have tried and tested this.

[Mastana420]

Oki thanks alot for the feedbacks. Wonder why Symantec technician suggested the other solution, and even said most of the customer choose this solution.

I woul love to know the comparison as far as performance is concerned, regarding these two solutions. and the performance using MSDP pools to achieve the same, as compared to utilizing Netbackup Appliances.

[RiaanBadenhorst]

Hi,

That statement is correct. Backup and restore operations occur within a domain. But as SymGuy-IT said, it is possible change the Master Server entry, effectively putting that client in another domain. The point being that its not an automated solution as with clustering.

[sdo]

You will also find that adding cluster aware storage underneath the cluster gets quite complicated and quite expensive.  If, at your site, no-one has implemented clustered storage before then you are in for an uphill struggle.  Also, consider that clusters and clustered storage are usually implemented on top of twin SAN fabrics and twin LAN networks which themselves are based upon diverse WAN routing.  i.e. there's not much point implementing a cluster and clustered storage for a HA cluster if the underlying infrastructure is not HA itself.  Technically it becomes even more difficult if your SAN storage arrays are not synchronous - i.e. if your storage arrays are asynchronous or are so far apart they they cannot achieve sub 10 ms committed synchronous writes at both sites then the performance will never really be acceptable.  Also, true HA also demands that no single component or functional area is ever driven above 50% utilisation - think about this... not only do you need two of eveything, you also need twice the bandwidth on each half/side/layer - so that if any path or component goes down then full demand can still be caterred for.  Anything less than this, then any failure could result in swamped/overloaded infrastructure which simply is unable to perform.

Cheaper and simpler, is one master at each site, and to use optimised duplication to replicate the backups that are most important to you.

It is definitely easy and do-able to restore from one NetBackup domain (i.e. one site) to another, just by adding the required server and/or media server entries on the clients.

[Mastana420]

Hi Thanks for the reply. 

Symantec dude i talked to said that performance vise, AIR can not be compared to optimized duplication. According to him AIR does generate alot of I/O processes, which could potentially require more powerfull master servers. According to him, that is not the case with optimized duplication. 

I dont quite understand what you mean by adding cluster aware storage underneath the cluster. As stated earlier, we are simply going to have 2 nodes at each site, active as Master servers in a cluster. The node in Site 1 will be the Active one, while the node in Site 2 being in passive mode, and taking over in case node in Site 1 fails. On top of that we will be having 1 appliance at each site. They will simply act as 2 different media servers in this domain. Only thing is that they will have optimized duplication enabled between them, so they are copy of each other.

Appliance in site 2 will have backups written to it, and duplicated to the appliance in site 1 by utilizing optimized duplication. And a similar setup for Appliance in site 1, where it will have its own backups written to it, which then will be duplicated to appliance in site 2, byt utilizing optimized duplication.  

If you are thinking about the master servers own disk, which it utilizes to write things such as catalog entries etc, the soltuion provided by Symantec technician includes SSD discs in each node which will be utilized to cahche things, to overcome any issues with latency between the nodes. 

[sdo]

Ok - I was thinking that you would have wanted your clustered master to be able to run and tolerate site loss - i.e. master node 1 at site A with SAN storage at site A -       OR     -      run as master node 2 at site B with SAN storage at site B     -      so my thinking was that you would have needed your SAN storage to be replicating between site A and site B.

Instead, it seems that you do not want to tolerate site loss, and simply want to be able to run master node 2 at site B using SAN storage from site A.    Trouble then is... what about disk IO (SCSI packet) latency from server at site B to storage at site A ?   WIll your SAN storage be FC or iSCSI ?

[Mastana420]

Hi sdo

The Whole purpose of the change is to be able to tolerate site loss by having 2 nodes across 2 sites in a cluster . But to achieve that we are going to be using FSS (Flexible Storage Sharing) which will utilize the internal disk of each cluster node in a RAID1 configuration. On top of that SSD disc will be utilized as caching disc to tolerate the latency between the nodes.

[sdo]

Looks good:

http://blog.proact.dk/tag/fss/

...especially my concern re where it says "What about performance penalties?
Writes will obviously have to be committed on both cluster nodes before the write is safe, which is why we recommend only 10GBit or InfiniBand solutions for this kind of setup."

.

The committed writes at both sites was my principal concern.  Looks like Veritas FSS has it covered.

.

What's the distance between the two sites?

[Mastana420]

We have fiber between the sites, and 10GB should not be an issue. Sites are less then 15km apart.

[sdo · Accepted solution]

Ok - assuming network team have considered latency and hops and possible QoS - then it might also be worth considering placing servers requiring low IO latency as near to the WAN ISL hop as possible, in an effort to reduce hop count - e.g. if the switches at each site carrying the LAN/WAN inter-site ISL have free ports - then maybe placing the servers utilising FSS on those "core" switches might result in lower IO reponse times than perhaps placing them on network edge LAN switches - but then maybe this also is not a problem if you have large wide ISL trunks between core and edge LAN switches at each site.

Something else to watch out for, maybe, is... if your inter-site fibre is not private dark fibre, then beware that some carriers splice different customers together, and sometimes the carriers will splice new customers in, or out, of shared inter-site carrier fibre - in which case it is usually best to use two different independent carriers with diverse routing - because these splicing "events / changes" can sometimes result in short-ish outages - and so also ensure that you have good communication with your carrier's "work teams" so that you can be informed of scheduled fibre works before they take place, and thus be sure that carrier A and carrier B do not engage in splicing events at the same time.

[Mastana420]

Thanks for the advise. We are already utilizing Netapp appliances in our network in the same manner, and they have huge amount of data replicating to each other without any issues, so i dont think network itself will be of an issue. We will involve our network team during this whole operation, so should make sure the best solution is chosen.

As for the other suggestion, i will makre sure to take that in concideration. Thanks.