P4500 MultiSite SAN and Upgrade

[ALgon]

Hello,

I have a few questions surrounding a multisite SAN configuration and an upgrade to version 10.5 we will soon be performing.  I’m still learning about this, so please bear with me.

Here’s an overview of our 4500 setup

We have two sites.

• We have a single Management group containing one cluster.
•  Four storage nodes are located at each site. 
• We have a failover manager node located at a third site.
• Site A is designated as the primary site
• Single subnet and VIP

My questions are:

General

• When a write happens from a server to a volume configured as raid 10, is it written to a storage node in each site simultaneously, or is it first written to the storage node in the primary site and then replicated to the secondary site?

• If the primary site goes offline, will the servers continue to access the data on the secondary site?

• If I have a two node windows application cluster with a node positioned on each site, should I assign each server to its corresponding site within the CMC?  What is the benefit of doing this?   I guess this depends on the answer to question one.

Upgrade

I’ve been reading the StoreVirtual Upgrade guide and I’m not sure on a few things:

• We have hyper V hosts that have the P4000 hardware VSS provider installed.  Do I need to uninstall this version before I install the Store Virtual Application aware snapshot manager on each Hyper V host?

• What’s the best way to confirm that all the servers accessing the storage have been configured with multipathing enabled?

 Thanks loads

[oikjn]

if the site is NR10, the writes are written to both sites before the write it confirmed to the initiator.  Generally if you are using the site aware DSM, the writes will be sent to the local site and then replicated to the secondary instead of the initial write going to the secondary site first which does help with access times, but you will always be limited by latency.  You talk about a windows cluster spanning the two sites, you should definitely assign each to their correct site in CMC as this will ensure they both have the fastest access to their data possible because they will then pull data only from their local nodes unless the local nodes go offline.

Upgrades from 9+ can all be done as online upgrades without uninstalls, but I think the DSM upgrades require reboots.

If the primary site goes offline, the secondary site will keep volume availability online, but interruption of access depends on how you have your initiators setup.  If you are using the HP DSM w/ microsoft, there is pretty much no interruption while the DSM reacts to the loss of the primary site.

Easiest way to verify all the nodes are connected correctly are to view the list of active connections in the iscsi initiator and also on the iSCSI session tab in CMC.

[ALgon]

Hi Oikjn,

Fantastic, thank you for answering my queries.

My only other question is  - The majority of our volumes are currently setup as RAID 10+2, and we'd like to claim some of this storage space back and convert these over to RAID 10.  My understanding is that the data would still be available even if we had a complete site failure and the only real difference is that we couldn't tolerate losing as many nodes compared to RAID+2.  Can these volumes be converted without any incurring any downtime, and is there any inmpact on performance during the conversion?

[oikjn]

the volumes can be converted online.  There WILL be an impact on performance because this requires the LUNs to be restriped, but you can control this impact by adjusting the managment bandwidth of the management group.  I generally turn this up as much as I can without killing my LUN latency and just keep an eye on it during the rebuild, but you can always just turn it down low and let the thing take a while to restripe if you aren't in a rush.

As for switchign to NR10 vs NR10+2, the difference would be that w/ a multi-site cluster you then end up w/ NR0 within the site and NR1 between the sites so that gives you site resilliancy, but if you lose a site AND THEN lose another node at the remaining site, your data is lost.  Generally, I would suggest you keep your truely mission critical data on NR10+2 and then move any data that you make the business decision to move other data to NR10 for any data that you would be OK with loss of data/availability should the unlikely event occur where you have a site failure and an additional node failure at the same time.  It may be unlikely, but it can happen so its more of a business decision to decide on the cost of downtime and data-loss vs the cost of storage space.

[RemyZ]

---

@oikjn wrote:

You talk about a windows cluster spanning the two sites, you should definitely assign each to their correct site in CMC as this will ensure they both have the fastest access to their data possible because they will then pull data only from their local nodes unless the local nodes go offline.

---

I think this depends on the bandwidth available between the two sites.

When the Windows cluster is an active-passive setup and you have plenty of bandwidth available, with low latency, I would not assign servers to sites. When assigned to sites, the servers will only access half of the nodes. I think you get more performance when the server can access all nodes.