EVA 5000 NetWare 6.5 Cluster Single Path?

[Guenter Knoechel]

We have a new EVA 5000 Array with 40 x 146 GB Drives in one Disk Group, 8 LUNs of 500 GB. Connected to both controllers are 2 x 16B FC-switches. We planned to attach 5 servers (Proliant DL360 with one FCA2110 FC controller) of a 10 server NetWare 6.5 cluster single-pathed to each of the switches.
Our problem in this configuration is, that servers on switch 1 see LUNs 1,3,5,7 and those on switch 2 see LUNs 2,4,6,8 only. Is there a chance to get this cluster working or do we need a second HBA and Secure Path for each server??

[Uwe Zessin]

Can you draw a picture of your fabric(s) and attach this?

All servers can only have access to a LUN through a single EVA controller at a time. It is not possible for half the servers to access a LUN though one controller and the other servers go though the second controller in case you have it set up this way.

[Guenter Knoechel]

It is probably the type of installation you are talking about. I have attached a graphical representation of the installation.

If this does not work, what is the problem and is there any workaround ??

[Uwe Zessin]

I see. Thanks for the picture.

For a single-path server without Secure Path connectivity you must make sure that the single fibre channel adapter in the server sees only one controller port, because the EVA reports its LUNs through all controller ports. This is done by setting up zoning on the fibre channel switch(es).

As I have already written, a server can only do I/O through one controller (both of its ports, at least), but the EVA _must_ report through all ports so that a multipath software (this is Secure Path, unless that functionality is built into the operating system) knows that there are additional paths it can switch to in case of a problem.

Let us assume that your servers 1-5 have access to the top controller, port-1. I guess that you have split your LUNs equally over both HSV controllers. Top controller = LUN 1,3,5,7 and bottom controller = LUN 2,4,6,8. If the servers 6-10 have been zoned so that they can see, for example bottom controller, port-2, this might explain the effect you see.

The server should see all LUNs, but NetWare might not use / show LUNs that it cannot do I/O to.

I don't see a way to really improve the situation without some investment. You _might_ be able to just put Secure Path on the servers and allow them to access the second controller port in its fabric. According to your drawing it is a port of the other controller, which is fine because this will protect against a controller failure. I say _might_, because I haven't done any work with Secure Path on NetWare, yet and I am not sure if this is a supported configuration - I am a bit short on time right know to check and I won't get to my desk for some days. Please ask someone with experience before you go that route.

What you, of course, can do is to put an additional fibre channel adapter in your servers and run with Secure Path. That way you can even survive a failure of a fibre channel switch without a server failover. It was a good decision to buy 16-port switches and not the 8-port ones ;-)

[Mike Naime]

Right now the week point in your design is the single HBA's in your servers. The EVA is designed to operate in Muti-bus failover mode. It was not designed for single pathed usage.

Since you only have a single path, you MUST be zoning out the other EVA port in your zoning configuration. Otherwise you would be seeing 2 paths to the storage! Because of this fact, the 4 connections on the EVA are gaining you nothing! You can only use 1 of them at a time. If you do not want to have the multi-pathing ability, then go ahead and plug all of the servers into the same SAN switch and Zone all of your servers to see the same port on the EVA. This is the only way that you will get all of the servers to see all of the LUNS at the same time. Go ahead and disconnect the other 3 EVA ports, they cannot do you a bit of good without the second HBA and Secure-Path!

The HSG and EVA controllers are Active/Passive in that the LUN is Active on only one port of one controller at a time. (Someone please correct me if I am wrong here) On a Multi-node VMS cluster, this is readily visible in that all of the Cluster members will be accessing the same LUN on the same port of the storage controller. When A path is lost, all cluster members will switch over to another path/port.


Getting the second HBA and Securepath will also solve this problem and allow you the redundancy that you already purchased to work correctly.

So to summarize:
Can you get this working? Yes!
Do you NEED securepath? NO!
Should you get Securepath? YES!
For the money that you have already spent on redundancy, I think you should spend the remaining $$ for a redundant environment.

How much will the downtime cost your company when something fails with your multiple single-point of failure design? Probably less than the HBA's, Fibre cables, and Securepath.

[Uwe Zessin]

Hello Mike,

I beleive there is a small error in your explanation. A LUN is 'online' to _both_ ports of a controller. The Tru64 Unix Version 5 operating system does use _both_ controller ports.

You should be able to verify it by entering the following command (I have no doubt that you are familiar with it) at the AlphaServer console:

>>> wwidmgr -show wwid -full

There should be four paths available. Two of them are marked 'connected = yes' (or something similar - I can't check right now). If you check the WWPNs they should lead to the controller ports where the LUN is 'online'. The other two connections with 'connected = no' are from the other controller.

[Mike Naime]

Uwe

While that is a True statement about the LUN being online to both ports, at the VMS OS level you only see one port active at a time.

Upon further checking, I can see that not all of the cluster members will have the same LUN on the same path. They are all on the same controller, but not necesarily on the same port.

Mike

[Uwe Zessin]

Mike,

you wrote: "The HSG and EVA controllers are Active/Passive in that the LUN is Active on only one port of one controller at a time."

I understood that as Active = Online on an EVA port as you didn't say anything about OpenVMS in that sentence. I recall that OpenVMS uses only one of the two controller ports - thanks for cleaning up the confusion.