Active IQ Unified Manager Discussions

Cluster Mode Question

UNDECIDED
4,939 Views

Hi all, I'm going through much documentation regarding Cluster-Mode and have a question I was hoping I could get some help with.

If you two 2240's in C-Mode, with three networks, management, cluster and data..(no cluster HA network as the 2240 doesn't have additional ports for it). Then the HA pairings will only be between the two controllers in the same filer....Does this mean that even if I have these two seperate filers clustered, if I were to lose one complete filer, i.e. both controllers on the same box I would lose the volumes on that filer since both controllers in an HA pair have gone down? Or would my data still be accessable on the other filer in the cluster? And if so would it be a seamless failover or would I have to invoke anything on the remaining filer?

thanks

1 ACCEPTED SOLUTION

mike_burris
4,939 Views

If I am understanding what you're asking for correctly, I do not believe it's possible.  The nodes in an HA pair are only aware of each other as it relates to disk ownership - only those two.  Meaning, unless I've missed something all these years, you cant have say 4 nodes (2 HA pairs) all cabled to the same loop/stack of disks.  So if that HA pair goes down somehow, any/all data on those disks is rendered unavailable until one or both of the nodes comes back online.

The only way I can think of to accomplish (I think?) what you're asking for is to setup a mirror to another node in the overall cluster but that's going to require disk capacity equal to the source aggrs/vols.  In the case of a load-sharing mirror, requests would be redirected to one of the "X" other copies for reads but you couldnt write data since the source would be offline.

View solution in original post

5 REPLIES 5

mike_burris
4,939 Views

As I seem to remember hearing/reading, the cluster network in //required// to be dual 10G ports so how are you proposing no cluster HA network?  There is only a single mezzanine port on the 2240's and the dual port 10G card is one of the two options available (other is FC ports IIRC).  The other nodes in the cluster dont have access to the disks on that HA pair so if it were to go down entirely (not sure how that'd happen honestly), clients will lose access to the data on those systems while that HA pair is down.

I should clarify here that I'm not well versed in C-Mode (been to some training but have never set one up, dont have on in lab or anything, etc) so anyone reading this that IS more experienced, please chime in and correct me if needed

scottgelb
4,939 Views

There is the Cluster which is all nodes in the cluster, then there are SFO (storage failover) pairs which are parts of the cluster.  In the case of a 2-node, the SFO is the same count as the cluster... if one node fails, the SFO partner will pick up all the volumes on the partner node (the same as 7-mode over the SFO interconnect to cluster the pair).  Over the cluster network (ethernet ports) all nodes still talk to each other for the quorum... in the case of a 2-node cluster we turn off epsilon (and turn back on when adding the or if ever adding the next SFO pair to the cluster).  For the 2220 it is also supporting cluster-mode off the 1GbE ports...although we might get the 10GB mezz card on the 2220 down the road.

UNDECIDED
4,939 Views

I have the dual port 10GB mez card option installed. So I will use these for the cluster network, but do I also need a separate cluster HA network if I want true active active clustering, so that if one filer goes down(both controllers) all data will continue to remain available without any user input?

scottgelb
4,939 Views

The sfo interconnect is internal in the shelf and is used for that kind of failover. The same as 7-mode.

Sent from my iPhone 5

mike_burris
4,940 Views

If I am understanding what you're asking for correctly, I do not believe it's possible.  The nodes in an HA pair are only aware of each other as it relates to disk ownership - only those two.  Meaning, unless I've missed something all these years, you cant have say 4 nodes (2 HA pairs) all cabled to the same loop/stack of disks.  So if that HA pair goes down somehow, any/all data on those disks is rendered unavailable until one or both of the nodes comes back online.

The only way I can think of to accomplish (I think?) what you're asking for is to setup a mirror to another node in the overall cluster but that's going to require disk capacity equal to the source aggrs/vols.  In the case of a load-sharing mirror, requests would be redirected to one of the "X" other copies for reads but you couldnt write data since the source would be offline.

Public