Lefthand/HP4330 design question

[jackjack2]

We have just purchased a Lefthand/HP4330 san, comprised of 6 nodes, with 3 nodes each placed in 2 separate buildings, configured in a stretched metro cluster, so network raid will distribute the data
across both sites, to all of the nodes.  Usable capacity is 18TB.  We are thinking of creating two luns to satisfy the storage heartbeat requirement; one small lun for iso's and the 2nd
storage heartbeat of approximately 300GB in size, and the remaining balance of the space, approximately 17+TB, as one large lun for all of the virtual machines.

The hosts are esxi/vsphere 5.1

Does anyone see any drawbacks to this plan?

Please ask questions if more information is needed to comment.

[a_p_ · Accepted solution]

With a 6 node Lefthand cluster, I'd consider to create multiple LUNs for better load distribution. Each LUN will have a managing controller, and with only a single production LUN you will concentrate traffic on one node and therefore loose the benefits of the Lefthand design. In this Mutli-Site cluster you plan, you will usually have 4 Managers (two on each site) and one Failover Manager (FOM), so I'd suggest at least 4 LUNs.

André

[jackjack2]

Andre,

Thank you for your response.  Is there documentation somewhere that you know of regarding best practices in carving up a metro cluster lefthand san?  So far I haven't found any. When I asked the VAR that came to install and do the initial configuration on the unit, all he said was that there had to be at least 2 for the vmware storage heartbeat.  If each lun has a managing controller though, then having multiples makes sense, I'm surprised he didn't mention this.

[a_p_]

When I'm back home, I will check whether I can find some documentation about best practices with multi-site configurations which may help answering your question. You may also want to take a look at the documentation provided with the system (the latest version is also available on HPs web site). Anyway it looks like the VAR was just thinking of VMware storage heartbeats and not about the storage itself!?

Btw. just out of curiosity, with the "stretched metro cluster" you mentioned, how are the sites connected to each other and what's the distance between them? In case of Ethernet, what's the roundtrip time for pings?

André

[ldelloca]

Hi,

it all depends if you are going to configure a proper multi-site storevirtual configuration or you are simply splitting part of the machines at the two different sites. I'm saying this because SotreVirtual has its own logic in balancing LUNs, and is not sometimes they way you would expect… look at this blog post for mor informations:

http://vmfocus.com/2012/12/03/gotcha-vsphere-metro-storage-cluster-vmsc-hp-storevirtual/

I'm not aware if the new 10.5 firmware has solved this or has some new mechanism to manage this, need to check in my lab. I do not have a multi-site deploy at hand, sorry.

Luca.

[jackjack2]

Andre,

I'm not sure of the exact distance between the sites but it is short, the two buildings are across a small parking lot from
each other, short enough to be within the range of the multi-mode fiber run between the two sites. The link is 1 gig. The ping response time between
the two esx hosts at each site is round trip min/avg/max = 0.205/0.237/0.270.

[jackjack2]

Luca,

Thank you for the link, it may take me a few reads to fully process it.  I'm not sure at the moment if we have a "proper" multisite configuration or not. 

[ldelloca]

No problem, it was in fact supposed to be a hint to check some configurations.

About the link connections, if you have a "campus cluster" with no difference in the speed link between intra-site connections and the inter-site connection, than you can also use the cluster as it is since the inter-link is not probably going to be a real bottleneck (depends on the amount of traffic) but anyway is not the best design to have VMs running in a site and using the storage from the other site. But as you can see from the link I gave you, there are problems with LEftHand to make it serve data from the correct site.

Luca.

[jackjack2]

Luca,

Does the need to insure site specifc storage apply if network raid 2 is being used?  When I asked the VAR installer where any luns created would "live" he said everywhere, that they would be spread across both sites.

It sounded to me as though the six nodes divided between the two sites would be seen as, and used as, one aggregate by the lefthand software.  But admittedly, the more I read, the more confused I seem to get.

[jackjack2]

sorry, meant to say network raid 10, 2 way mirror.

[ldelloca]

Yes, the explanation about the LUN spreading is completely correct, if you use Raid-1 (I think you were meaning this, not 2…) basically each side has a copy of the LUN itself. What changes (and that's the reason I suggested that link) is that at a given time only one of the P4000 is the gateway of the LUN, the others are only syncronyzing the copy of the LUN they have.

This means, in a flat scenario with LeftHand, at some point the LUN could be exposed by a storage node in site 2 but accessed by an ESXi server in site1. This is not a real problem if the inter-link between the two sites has the same speed as he local network, like you confirmed. You will only see an increased amount of traffic in the inter-link both from ESXi accessing storage across the link , and the several storage node synchronizing each other.

Luca.