Version 11.0 Announced, Finally - Capacity Increase and Tiered-Storage

[oikjn]

lol.

wonder what the problem is.  Only thing I can guess is that they have some sort of problem doing an UPGRADE of an existing system since the clean install has been out for a while.

Anyway...  beyond the fact that 11.0 isn't even fully released yet, its a dud on release anyway without support of UNMAP and presumably the never-to-be-seen deduplication feature as they decided to make a Storeonce VSA :-/  .  AO will be nice, but anybody who really cared about that could have done it already at the hardware level with LSI's cachecade a year or two ago.  Maybe AO is "better" than Cachecade, but now you are talking about a % here or there.  I just hope 11.5 is announced soon and at least catches up StoreVirtual to be back on par for features of all other modern SANs. 

[AuZZZie]

I agree that, as usual the release doesn't really bring anything to the table feature wise (except SMI-S.. Finally). The reason I'm getting annoyed however, is that Hyper-V 2012 R2 DOES bring something to the table, and it isn't officially supported without StoreVirtual 11. 

[oikjn]

that is very true.  I wasn't thinking about that as I only upgraded some VMs and not my hosts that connect to the SAN, but that is deffinitely a problem for those who are ready for 2012R2.

[Andre van den Berg_1]

There are now two 11.00 patches available but still no version 11.00 for upgrade.

[AuZZZie]

All HP seems to reply is that StoreVirtual 11 has been on the web since "blah blah". They don't seem to understand their own product that for existing deployments (production) that is of absolutely no help and doesn't allow you to upgrade without the CMC upgrade path. 

[5y53ng]

Howdy, been away for a while... 

So has anyone actually implemented the tiered architecture with 11.0?  I just had several Dell R810s dumped in my lap that have been collecting dust. I was considering using these to create a test SAN with 11.0 and load them up with a mix of 15K SAS and SSD to see what the performance is like.

Do you need to create a physically separate RAID group of SSD to implement the auto-tiering?

[Bart_Heungens]

Hi,

I have it running already some time inside my environment.

What you will do is create indeed 2 types of VMFS datastores with different underlying disk technology (R5/R1 or SAS/SSD...) and that you will put your VMDK files on those different datastores...

Afterwards inside the CMC you have to define those 2 different VMDK files as 2 different tiers...

I will be posting this week (if time permits, lots of information collected at HP Discover last week) on my blog on OA on StoreVirtual...

www.bitcon.be

Kr,

Bart

[5y53ng]

Bart,

Thanks for the reply, sounds great! 

[twg351]

Looks like some have been able to get OS 11.0 up already, am I to understand two keys things about OS 11.0:

(1) OS 11.0 is the first version to allow Windows 2012 R2 to work with the SAN (iSCSI initiators, MPIO, etc.)

(2) Windows 2003 will not in any way be able to connect to the SAN's anymore (even a Windows 2003 server trying to connect a single volume)?

Sounds crazy, but I do still have some legacy W2003 systems using SAN volumes.  I am also looking to built a Windows 2012 R2 HyperV cluster early 2014.  

I get the feeling I can have one or the other?  Anyone confirm this?

Tom

[oikjn]

I haven't tested because I can't upgrade to v11 yet, butI believe the issue is the DSM for 2003 is no  longer supported.  I don't know why you wouldn't be able to use the MS MPIO option, but I would assume you would have to lose some of the benefits of the HP DSM.  It might work, but all in all at some point support has to end for these old OSes.  I"ve still got some legacy things running on 2k3, but at this point they are all VMs and not directly attaching to my SAN so I think I"ll be ok.

[AuZZZie]

CMC Upgrade path is delayed yet again now to Jan 15th according to the StoreVirtual 11 download page. That's only like the 4th or 5th date change though so take it with a grain of salt. 

I'm officially done with HP storage. Soon as next years budget allows I'll be moving to a new vendor, and I'll be ceasing to recommend HP storage to our clients. Love their servers, but their Storage team all need to be fired. 

For anyone wondering though, I upgraded my Hyper-V cluster to 2012 R2 and used the v11 DSM while keeping my SAN/IQ nodes @ 10.5. All seems to be working. So if you're like me and have given up on HP ever releasing it, you can still move to 2012 R2 and remain stable. 

[Bart_Heungens]

Hi,

I do not know why the update is delayed again until january, but to say that you will stop buying HP storage for this does not make sense... At least for me...

Version 11 is already available for download via another way and I was able to update my Lefthand environment already a few weeks ago to 11.0, although not via the CMC way...

But if you really really really need/want 11.0, know that there are other ways...

Kr,

Bart

[oikjn]

well I wouldn't not buy because of this, but I hate to say it, but the pattern of advances of this system going back to v8.5 has been a constant and increasingly large lag in features of modern SANs.  What does v11 get you for the hardware nodes?  SMI-S support? "increased performance"?  2012r2 support?  Those aren't "features", those are basic requirements of a SAN.  The VSA was innovative 5 years ago and I"m sure they could make it innovative again if they cared, but either they care and don't communicate it and have terrible results or they don't care and will let it die.

I decided to stick with this SAN this year, but come next it may be another story if there isn't a clear roadmap for the future of this product.

BTW, I have a few VSA nodes and would like to take advantage of AO.  How did you upgrade?  Just install a new node and do a node exchange?  X-mass is a good time for me to play around with this stuff. 

[Bart_Heungens]

I understand your opinion, and partly I had the same feeling... Last week at HP Discover I could speak quite some HP people including Lefthand representatives, and I cannot tell you all the details (under NDA) but I can guarantee you that 2014 will be a great year... Believe me...

BTW at HP Discover there was a new model on demo that will be officialy launched/presented 1 of the upcoming weeks. AO not only on VSA but now also on a physical node... So yes SSD and SAS inside 1 node... I am currently writing a blog article on it and will publish it soon on my site...

Yes I did it thru the node swap feature... Worked fine, not ideal but shows me all the features of the product. Did already a lot of upgrades thru the CMC, sometimes it is fun to do it the other way...

[oikjn]

thanks.  I really like the VSA line and it does have a ton of potential.  I do sincerly hope HP does condinue to actively develope it.  Nice to at least hear the rhumor that things are comming in 2014.  I just wish they would publish a timeline or communicate something to us...  heck, they have this forum, they could post something here or have a representitive here post something.  Is there even a person on the team on this board?

[Delta69]

Hello,

I have 2 questions regarding VSA 2014 / LeftHand 11.0. Perhaps someone could help ?

- Can you confirm it's possible to go beyond the global 50TB limit buying additional licences (with additional hardware for these licences, of course...), so as 100 or 150TB could be seen from a unique management console ? (In other words, I do not want to split the architecture into several separate 50TB vSANs managed separately !)

- What about the global performance when all the data is spread among multiple identical servers ? Is global performance increasing in adding servers and storage, or is it better to have  as less server as possible, but choosing the most powerful ones ? (I'm not dealing with data security, just performance...).

Thanks for your feedbacks !

[a_o]

AFAIK, there's no 50TB global limit.
Version 11 intoduced a new 50TB VSA license.

There's a limit of "Up to 3 StoreVirtual VSAs per cluster" - i.e. a cluster can have 150TB of raw storage in VSA.

The current physical appliances support up to 1536 TB per NSM. (i.e. BL series based - 4630sb  w/ 32 D2700 expansion chassis)

So, using the VSA limits, StoreVirtual can scale up to 4608 TB per cluster using physical appliances. (i.e. up to 800 SFF SAS disks per cluster)

I've read about an unofficial 10 node limit in a cluster. But I've also seen an HP sponsored ESG report showing a configuration with 30 nodes in a cluster servicing 100,000 Exchange users.

However, there's no limit on how many clusters that can be configured in a single management group.

[a_o]

Regarding performance, it depends ...
I don't fully understand what you mean by "adding servers and storage" But I'll give it a try.
Like any I/O based issue, you can always throw more spindles, CPU and I/O bandwidth at it.

StoreVirtual scales well in the sense that with it's clustering methodology, multiple nodes in the same cluster can service requests for data in the same LUN from different hosts at the same time. Coupled with MPIO you can get even better performance. So, yes more storage nodes will increase performance.

In any case, there are so many ways to skin this cat and YMMV.

With the new Gen8 based NSMs with full SSD and 10GBE/FC or the newer ones with AO, you'll have all of your performance requirements met I think. Granted that it's not exactly 3PAR SSD or Pure Storage.

My take is that if you need the best performance from StoreVirtual, go with VSA vs NSM.

We started with P4300/P4500 NSMs. I wasn't happy with the performance -especially with SQL server.
Eventually, I converted our setup to VSAs, using 10GBE, BL I/O accelerators and a mixture of MLC SSD and 15K/10K SAS drives. The cherry on top for going with VSA's is that when needed, we can do HW Raid 1 or 10 on the physical drives as opposed to being limited to 5/6/50. Makes a difference in the write performance.

In all, we have 16 blades running one to two VSAs each depending on what kind of data is being served.

I'm much happier with the performance we're getting.

[oikjn]

@a_o - I think the 3-node limit is w/ the new smaller VSA license, I think there isn't any change to the limit recommendations on the 10 or 50TB nodes.  HP gives its recomendations on the number of nodes/clusters per management group and I can't remember them offhand.

One thing to keep in mind is there is a limit to the total number of iscis connections, so depending on how you connect this can become an issue as the scale goes up.  At a point you can't use full active MPIO because of the iSCSI connection limits, but w/ 50TB VSAs, that issue is a bit mitigated as you can put more data onto a single node.  ( I think the issue starts to crop up typically after a cluster gets more than 8 nodes in it).

[Bart_Heungens]

Hi,

The 1TB is a free license that comes from now on with the most popular HP Proliant Gen8 servers. There is indeed a max 3 node count in a cluster.

The other (non-free) licenses of 4, 10 and 50TB are not limited as such... Inside the CMC, there is a best practice analyzer that will tell you about the max nodes in a cluster and management group. Depending of who you ask at HP it will be somewhere between 16 and 20 in 1 cluster...

kr,

Bart

[Delta69]

Thanks a lot guys for your feedbacks... I really appreciate... It'll help for sure.

By the way, Merry Xmas !

[oikjn]

bart, had a quick question I PMed you... about the practical side of if you can have a split v11/v10.5 management group.

[Bart_Heungens]

check mail!

[oikjn]

just an FYI.  I was able to delete an existing VSA and exchange it with a clean installed v11 into a v10.5 cluster, so v10.5 and v11 can co-exist in the same cluster :).  For anybody who tries this, make sure you stop the manager on the VSA before you destroy it.

[Northwoods]

The 10 node limit is what HP tests the software against. The engineering maximum was 32 nodes (not-supported) awhile back, not sure if this has increased or not. The problem with large clusters is the amount of background chatter you get...think of all the backend replication traffic for a multi-site cluster with 16-nodes at each site.