Buying new SAN, any one with experience Veeam / Compellent?

[perkristian]

Hi!

Today in our 6 hosts vSphere 5 environment we are running our Veeam backups to a Datadomain D610 replicating of site and our Replicas to a IBM DS3512 SATA SAN onsite.
We are currently in the proses of acquiring a new SAN. We are looking at either a iSCSI DELL Compellant or buying another traditional FC IBM DS3512 SAN.

So my question is: Have any one any experience with backup and replica speed traditional SAN vs. Compellent?
As I understand VMware lets Compellent handle all snapshots so this should theoretically speed up both backup and replica time as form our Veeam logs I see that the attach/detatch delete snapshot proses takes a long time today.

Thanks in advance for any thought around this!

Brg, Per Kristian

[dellock6]

Hi Per,
I have no direct experience on Compellent, I read something about the Data Instant Replay feature but I do not know if it's something near to a vCenter snapshot so it can be used by Veeam (or other snapshot-based tool for VMware). I'm not sure anyway it's done at the VM level or at the LUN level like may other storages. The only storage I'm sure works at the VM level is Tintri...

By the way, Compellent is really not comparable with an IBM DS3500, completely different level (which means, Compellent is way better...).

Luca.

[tsightler]

I've seen Compellent in the field a couple of times and I can say that with ESXi 5 and VAAI enabled for the Compellent that snapshots are quite fast, perhaps the fastest I've seen on any storage (we were seeing snapshot removal times that were basically 1 second, and that was on a pretty busy SQL server). I was impressed with this combination and Veeam works great. These environments were using a Veeam physical server with direct access to the SAN via FC. We configure the Compellent to present the VMFS LUNs as read-only so no chance of any corruption.

[dellock6]

Tom,
thanks for sharing! but are those LUN snapshots or VM snapshots?

Luca.

[tsightler]

VM snapshots.

[dellock6]

This sounds really cool!
Is it something I see at the storage level, or are they shown as standard snapshots inside vCenter?

Thanks.

[perkristian]

Thanks for all your replies!

Looks like this definitive could speed things up regarding to snapshot times.
I'll keep you posted about how it works with Veeam if we decide to go for the Compellent solution.

Thanks again!

Brg, Per Kristian

[tsightler]

They were just regular VM snapshots. I think the reason they were so fast is due to the way that Compellent leverages their storage profiles. I'm not a Compellent expert, but they appear to use a storage tiering technique where, based on your storage profile settings, all writes are always made to the higher tier (in this case I think it was SSD) and then moved to the lower tiers as the data ages. Because snapshots by they very nature are only writes, they are in the highest performance tier and thus can be committed quickly. That, paired with their VAAI to reduce the overhead of cluster reservations and metadata updates and you can see some amazing performance.

[jwaynejones]

We use a Compellent SAN in our production environment, and it's been a good experience overall. Looking at your original post, though, I think you might not understand the SAN snapshots. VMWare doesn't offload the snapshots to the SAN, it's an additional layer. All VMWare related snapshots are still taken in VMWare, and are completely independent of the SAN snapshots (or "Replays"). You use the SAN snapshots for a handful of features, but they do not replace VMWare snaps. The Compellent does use an automated tiering system (Data Progression), which is overall good, but I would warn against buying into their "let the SAN handle it all" mentality. If you have systems that require high throughput, you're going to want to apply a storage profile that keeps them in Tier 1. The best practice is to have the active data (writes for the most part) in a RAID-10 configuration, and the replay data (read only) in RAID-5. We've seen very good performance with this setup, but you also save a lot of space by not having the entire LUN in RAID 10. You can also run SAN to SAN replication via the Compellent if that's something you're looking into. It's actually very effective, you just have to make sure you understand SAN replays. Their replication is also supported in SRM 5.

It's a great product, just make sure you get a good understanding on Data Progression and SAN replays.

[joergr]

As my previous speakers already mentioned, a vmware snap operation is NOT being offloaded to ANY san at this time (yes yes hardware assisted locking is kicking in but thats not the main reason. maybe vmware will add that to vaai in the future sometime). The main reason why compellent may look fast during snapshot creation/consolidation is their fluid data idea of moving all the hot writes to the fastest drives with the most powerful raid level (raid 10 with ssd rocks, i can tell you). BUT like jwaynejones told you, don´t count on the magic super logic of fluid data self-configuring-machine. Monitor your system closely. Configure it according to YOUR needs.

I´d also suggest to take a close look at DELL-Equallogic. Let´s take the new 6110X, it´s a 24*900 GB 10K SAS System with 10GB Ports (Copper and Fiber) for a very good price, configured as RAID10, this thing performs like crazy, it screams.

Best regards,
Joerg

[Stephenv]

We use a Compellent SAN; I must say I do love this thing.

My Veeam backup server connects directly to the Compellent SAN in read only mode.

VMware handles the Snap shots not the Compellent. (Unless you do some sort of RDM mapping, I don’t)

Veeam reads only the Change block data so it’s quite quick. Also the reads go from the SAN to the Backup server and not through the ESX host.

As for the replication, I don’t know much, I do know the SAN to SAN replication is quite good. (Compellent to Compellent)

Stephen

[dellock6]

So basically everyone using Compellent are confirming VM snapshots are still made by vCenter as usual, and Replays are LUN snapshots. Thanks for clarification, this what I was suspecting too...

There is a storage that actually do snapshots at the VM level, and is Tintri:
http://www.tintri.com/products/features/
as you can read at the end of the page. I would like anyway to validate this feature by hand with a demo from them instead of only reading their website.

Luca.

[bhwong]

What is the diff between Dell compellent vs equallogic?

[dellock6]

I've not worked so extensively with Dell storages, so do not keep my words for granted...
For what I know, EQL is a SAN based only on iscsi protocol, that leverage redundancy by adding storage nodes to the same cluster, like HP lefthand does. Basically, you do not have redundancy from a single storage unit (unlesss for its internal raid on disks) but by adding more nodes you get a network raid at the end, and every nodes adds its compute power to the total amount available, both from IO and cpu perspective.
Compellent has a more classic design in the sense it has storage controllers and disk shelves behind them, you can scale by having multiple storage controllers and more disks on every controllers. It's on a higher level because it can export luns in multiple formats (FC, iscsi....) and has automatic tiering (different performance layers of disk, and it moves cold data from higher level to low level automatically).

Luca.

[Gostev]

What is the diff between Dell Compellent vs Equallogic?

Price

[J1mbo]

Adding nodes to a storage cluster on EqualLogic adds load balancing and capacity, but not redundancy since that's handled within each node.

For example you could have SAS and SSD nodes in an EQL group, each has node has a single RAID policy (10/50/6 etc) and dual controllers. Everything is iSCSI based. Once LUNs are defined, optionally a RAID preference can be set or left to automatic, in which case the EQL will scan access patterns and migrate blocks accordingly between the storage classes. But, each node isn't expandable so capacity expansion is by adding more nodes.

Compare to HP's P4000, that's very different as each node is really just a Proliant server with lots of disk bays, so because they then lack dual controllers they have two complete notes and as a result there is massive overhead in terms of physical to usable disk storage (and power consumption with it).

If deploying EqualLogic on vSphere 5, be aware of this and this (some care in configuration needed).

Hope that helps.

[johnlockie]

To be fair, AFAIK Compellent is a multi-protocol SAN/NAS, while EqualLogic is a single protocol SAN (iSCSI)...so it's no surprise one is more expensive.

What I am curious about in this thread is two things - since we seem to have the storage discussion going:

1. Has anyone experience with both Compellent and NetApp or EMC in a virtual environment? If so, candid thoughts are appreciated...along w/ protocol experience (CIFS/NFS/iSCSI/FC)
2. Are any of you on Compellent using VDI in your environment? If so, how do you address the read bottleneck presented with VDI? Because I am reading that you can use flash storage in combo with storage profiles to prioritize write, but this is backwards to my SAN experience. From my experience reading can be the issue, which is the reason for flash cache. NetApp has their PAM cards (performance accellerator), and I believe 3Par and EMC both have solutions for this too (cache accelleration).

Thx
j

[J1mbo]

....as do EqualLogic, the idea there is that you buy an SSD shelf and the group migrates the VDI stuff onto the SSDs ready for the Monday morning rush (or whenever else it might be). Actually they have a hybrid model so one box can do that, from the blurb the EqualLogic Array Software automatically tiers data to optimize pool performance at the sub-volume level-delivering the right data to the right place at the right time.

[creativeTechie]

I ran across this thread and I am hoping someone can answer a quick question. If Veeam reads the Compellent SAN via direct access, will that mess with the Data Progression system?
I would love to hear anyone's experience with this because we have been using Veeam for a long time but recently purchased a Compellent SAN.

Thanks,
Matt

[foggy]

While I had no experience with this, to my understanding, since Data Progression is a sub-LUN tiering solution, it should not produce any issues to direct SAN access.