HPE EVA Storage

Continuous Access EVA Zoning

 
Heather O'Neil
Advisor

Continuous Access EVA Zoning

Hi all,

We deployed HP CA over last weekend, but ran into some interesting issue. I am thinking it is zoning related, but would like some feedback. Here is our setup.

We have a primary site in Lancaster, PA and a DR site in Las Vegas, NV. We have a dedicated T3 between the sites with an average RTT of 55ms. We deployed CA using asynchronous replication. We have (4) Cisco MDS 9506 switches, (2) per site. The long distance connectivity is done via fcip tunnels using IPS modules. We have one gigabit interface per switch (slot 1, port 1 or g1/1/).

We created an FCIP tunnel from switch1 in Lancaster to switch 1 in Vegas. We also created an FCIP tunnel from switch 2 in Lancaster to Switch 2 in Vegas. At each site we have (2) HP EVA4400s, both running the latest code, all switches are also running the latest firmware.

Since we don't want to merge the fabrics, we are using IVR. After creating the FCIP tunnels, we saw a lot of VSAN segmentation, since the VSANs tried to merge. Which I would think it is expected since the VSAN IDs are the same across each fabric, but the zone configurations differ. Am I correct?

The EVA resides in fabricA VSAN100 (ports 1) and FabricB VSAN200 (ports 2). This applies to both sites. We used IVR to configure IVR zones as follows:

Command View Management Zones

Lancaster
FabricA
Zone Name: CommandViewMmgmtA
Members: CVLancaster_HBA1, VegasEVA_C1_P1, VegasEVA_C2_p1

FabricB
Zone Name: CommandViewMmgmtB
Members: CVLancaster_HBA2, VegasEVA_C1_P2, VegasEVA_C2_P2

Vegas
FabricA
Zone Name: CommandViewMmgmtA
Members: CVVegas_HBA1, LancasterEVA_C1_P1, LancasterEVA_C2_p1

FabricB
Zone Name: CommandViewMmgmtB
Members: CVVegas_HBA2, LancasterEVA_C1_P2, LancasterEVA_C2_P2

IVR Replication Zones

FabricA
LancasterEVA_C1_P1
VegasEVA_C1_P1

FabricB
LancasterEVA_C2_P2
VegasEVA_C2_P2

Here is my issue and questions.

The command view server at each site can see the EVA at the other site. However, the response time is very bad. Command View works fine when viewing resources of the local EVA, but takes forever when viewing the remote array. Sometimes it takes over 5 minutes to refresh and we get the communication lost page.

Could this be a zoning issue?

Does the zoning look correct? Are two fcip tunnels supported through a single T3 line? I have looked everywhere and can't find an answer.

When you dedicate controller ports for replication, should those ports be zone out to separate from host I/O traffic? It seems like that would make sense.

One more thing I should mention. While troubleshooting I noticed that if the fcip tunnel for fabricB was shutdown, the command view servers couldn't see the remote EVA. When it was brought back up, it could. Since fcip tunnel for fabricA was never brought down, shouldn't the Command View servers still see the remote EVAs?

Any feed back is greatly appreciated.

Thanks!
6 REPLIES 6
Uwe Zessin
Honored Contributor

Re: Continuous Access EVA Zoning

Heather, you must first adjust your zones for the EVA-4400.

There must be exactly one adapter port and one EVA port in one zone. Do not put both EVAs in one management zone - this creates a 'hidden' replication link. Better create an additional zone for CA.

The EVA uses internally the IP addresses 10.0.0.1-3 and although it is not officially said: the wording in some documents (and experiences from other users) imply that the management module (also called the WOCP) exposes this internal communication via its external LAN link !!!

Do not put any other device with those addresses (this includes another EVA-4400 WOCP) onto the same 'LAN segment'.
.
Víctor Cespón
Honored Contributor

Re: Continuous Access EVA Zoning

I've not done a detailed analysis of your config, but here are a couple of quick toughts:

Do not include two ports of the same EVA4400 on one zone. This usually leads to external Command View losing access to the EVA.

Each controller must see both controllers of the other EVA. Without this some commands may fail.

Yes, ports used for replication should be on independent zones.

Command View only talks to master controllers, take that into account too.
Uwe Zessin
Honored Contributor

Re: Continuous Access EVA Zoning

Good point - here is an example for one management server:

Fabric A:
Zone: CVLancaster_HBA1, LancasterEVA_C1_P1

Zone: CVLancaster_HBA1, LancasterEVA_C2_P1

Zone: CVLancaster_HBA1, VegasEVA_C1_P1

Zone: CVLancaster_HBA1, VegasEVA_C2_P1

Fabric B:
Zone: CVLancaster_HBA2, LancasterEVA_C1_P2

Zone: CVLancaster_HBA2, LancasterEVA_C2_P2

Zone: CVLancaster_HBA2, VegasEVA_C1_P2

Zone: CVLancaster_HBA2, VegasEVA_C2_P2
.
Heather O'Neil
Advisor

Re: Continuous Access EVA Zoning

Uwe,

Thanks so much for your quick reply. You mention that I must adjust my zones first and to make sure that there is one adapter port and one EVA port per zone. You also mentioned to not put both EVAs in the same management zone. I double checked our zoning and while we don't have it zoned as you mentioned (one hba port & one EVA port), the management zones do not include both EVAs in the same zone. See below.
CA Management Zone Lancaster (Local) Includes Local EVA

Lancaster
FabricA
Zone Name: CommandViewMmgmtA
Members: CVLancaster_HBA1, LancasterEVA_C1_P1, LancasterEVA_C2_p1

FabricB
Zone Name: CommandViewMmgmtB
Members: CVLancaster_HBA2, LancasterEVA_C1_P2, LancasterEVA_C2_P2

CA Management Zone Lancaster(Remote) Includes remote EVA

Lancaster
FabricA
Zone Name: CommandViewMmgmtA
Members: CVLancaster_HBA1, VegasEVA_C1_P1, VegasEVA_C2_p1

FabricB
Zone Name: CommandViewMmgmtB
Members: CVLancaster_HBA2, VegasEVA_C1_P2, VegasEVA_C2_P2

CA Management Zone (Local)Vegas includes local EVA

Las Vegas
FabricA
Zone Name: CommandViewMmgmtA
Members: CVVegs_HBA1, VegasEVA_C1_P1, VegasEVA_C2_p1

FabricB
Zone Name: CommandViewMmgmtB
Members: CVVegas_HBA2, VegasEVA_C1_P2, VegasEVA_C2_P2

CA Management Zone Vegas(Remote) Includes remote EVA

Vegas
FabricA
Zone Name: CommandViewMmgmtA
Members: CVVegas_HBA1, LancasterEVA_C1_P1, LancasterEVA_C2_p1

FabricB
Zone Name: CommandViewMmgmtB
Members: CVVegas_HBA2, LancasterEVA_C1_P2, LancasterEVA_C2_P2

When you say that I have to create zones that only include one HBA port and one EVA port, is this something that needs to be done just for the CA Management zone or for all host zones? I have always zoned hosts and EVA as follows:

FabricA
HostZone: Host1_HBA1, EVA_C1_P1, EVA_C2_P1
I remember there was some advisory with the EVA4400 that HP recommended host zoning of one EVA port per HBA port like this:
FabricA
HostZone: Host1_HBA1, EVA_C1_P1
HostZone2: Host1_HBA1, EVA_C2_P1
I didn't think this was necessary on the 09522000 code. I could be wrong.

Does the replication zones I have configure for the EVA4400 array look ok?

FabricA
LancasterEVA_C1_P1
VegasEVA_C1_P1

FabricB
LancasterEVA_C2_P2
VegasEVA_C2_P2

After doing further research, I seem to think the replication zones using a single T3 line should look more like this:

FabricA
LancasterEVA_C1_P1
VegasEVA_C1_P1

Basically, nothing gets replicated through fabricB.

Thanks again for all your help !
Heather O'Neil
Advisor

Re: Continuous Access EVA Zoning

vcespon,

Thanks for your reply! I will go ahead and re-zone the switches ensuring that there is only
one EVA port per port for the management zones. Just as Uwe has described. Thanks Uwe!

You mention that "Each controller must see both controllers of the other EVA. Without this some commands may fail."

The way I currently have the replication zones configured are as follows:

FabricA
LancasterEVA_C1_P1
VegasEVA_C1_P1

FabricB
LancasterEVA_C2_P2
VegasEVA_C2_P2

As zoned above, doesn't that allow the controllers to see each other? or should it be more like this:



FabricA
LancasterEVA_C1_P1
Lancaster_EVA_C2P1
VegasEVA_C1_P1
VegasEVA_C2_P1

FabricB
LancasterEVA_C1_P2
LancasterEVA_C2_P2
VegasEVA_C1_P2
VegasEVA_C2_P2

As zone above it would ensure that all controller ports can see each other, but it would also mean that all EVA controller ports will participate in the replication. This is something we do not want as we want to dedicate most of the ports for host I/O.

Since it is only a 45Mb pipe, my thought was to simply use the minimum number of required EVA ports for the replication. Since all DR group members are owned by the same controller, we want to include ports from both controllers in the replication. This way would could load balance the DR groups across both controllers. Does the replication zone I wrote above look correct?

Are there any issues with the replication zones being IVR zones?

Best practices indicate that host I/O should be separated from replication traffic. Does this mean that any ports used for replication should be removed from zones that use those ports for host I/O? For example, if a zone includes the members: (Host1HBA1, EVAC1_P1, EVAC2_P1) we want to use port EVAC1_P1 for replication, should that port be removed from the host zone?

I may be totally wrong here, but what is the minimum number of ports supported by CA? I was under the impression that I could do with as little as one port from each controller
(i.e. LancasterEVA_C1_p1, VegasEVA_C1_P1)

Thanks in advance!
Joseph Salzer_3
Advisor

Re: Continuous Access EVA Zoning

Heather,

Have you been able to find a solution to your problem? We are running in to an interesting issue as well. We have our command view servers zoned as others have described. Basically the command view servers only see one controller port per zone, per hba. Our replication zone looks like this:

LocalEVA_Ctr1_port1
LocalEVA_Ctr2_port1
RemoteEVA_Ctr1_port1
RemoteEVA_Ctr2_port1

When we try to replicate, we get an "Attribute Mismatch" error. Can't find anything on it. Anyone have any ideas?

I should probably open another thread, but I think our issues may be similar.

Thanks,
Joseph