MSA Storage
1748063 Members
5593 Online
108758 Solutions
New Discussion юеВ

Re: MSA1000, Zoning and Redundant Switches

 
SOLVED
Go to solution
CA1281521
Occasional Advisor

MSA1000, Zoning and Redundant Switches

Hi,

I have the following setup:

4x DL580G2 With 1xQLA2340 each
2x DL360G3 With 1xQLA2340 each
1x MSA1000 with Redundant Controllers and one 8/2 switch

The 4 DL580G2 nodes are used for an RH Oracle RAC setup and the 2 DL360G3 are a RH Cluster with NFS and MySQL with RHCS.

I have no SSP in the controller or Zoning in the switch.

Everything works fine, but sometimes (1 out of 10) when I reboot one of the machines on the fiber loop, all the machines loose the connection on the LIP Reset and I have to reboot all of them to make them work again.

The same thing (worst, 8 of 10) happened to me in a similar setup but with a Brocade switch and Infortrend Storage system. I solved it with zoning.

Will zoning help me here also?

How can I access the switch configuration?

With the same setup I also want to add a redundant switch. Given that the hosts thay connect to it have just one interface to the switch, I plan to connect half of them to one switch and the rest to the other. If one switch fails at least half the machines would be alive.

Am I right? How will a controller failure affect this system?

Thanks a lot!
11 REPLIES 11
Uwe Zessin
Honored Contributor
Solution

Re: MSA1000, Zoning and Redundant Switches

Oscar,

you write that you have redundant controllers...

What type of multipath software is in use on your systems? The problem with Linux is that a boot can cause a logical disk failover to another controller on the MSA1000. As the MSA1000 is working active/standby it means that all logical disks will fail over.

It sounds like you have some systems that do not share all logical disks. In that case, SSP is *definitely* a good idea! Imagine a destructive command against a wrong logical disk and your data is gone!!

Separating the fibre channel adapters by zoning is another good idea as it will minimize the effects to the oder adapters when one adapter enters or leaves the fabric.


Now, from your description it appears that you have an invalid configration! A single adapter must only have access to one MSA1000 controller at a time and if you have only one fibre channel switch, you need to do that with zoning.

You cannot connect half of the systems to one switch and the other half to the second. Well, you can, but it will not work ;-)

The MSA1000, like I wrote above, works in an active/standby mode, which means that all logical disks are served by only one controller. A failover of one host means a failover of ALL logical disks.

You need to put redundant fibre channel adapters or adapters with dual ports into your servers and, again, you need some kind of multipathing software.

When a controller fails, the multipathing software 'knows' that another path exists to the standby controller and it will active this path.
.
CA1074530
Advisor

Re: MSA1000, Zoning and Redundant Switches

Hi Oscar,
I have been advised by Brocade that best practise is to create a zone per host storage connection.
Reason given to me was due to the SCSI protocol, and would make sense given the description of your problem. Basically when you reboot a server or device on the fabric, a SCSI reset command is issued to all devices on the bus, and can cause devices to be dropped from hosts etc. This is why they reccomended to me to zone per connection, or at least per device. eg: best is as below for multiple devices:
Zone1: Server1, MSA1000
Zone2: Server1, Tape device
Zone3: Server2, MSA1000
Zone4: Server3, MSA1000

or at least:
Zone1: Server1, MSA1000, Tape device
Zone2: Server2, MSA1000

I have configured my fabric as per the first example and have noticed an improvement. I used to loose tape drives constantly. When I zoned per device, I have not had a device disconnect from a server.
Uwe's comments are also quite good and valid.

What type of switch do you have? If it is Brocade then you can use the serial interface.

Dean.
CA1281521
Occasional Advisor

Re: MSA1000, Zoning and Redundant Switches

Hi Uwe,

I'm not using multipath software, right now I have both controllers in the same enclosure.

I see you point on the failover and the controllers. I need multipathing software for a controller failover to happen.

When you say I can't connect hosts half and half, is it because each switch is connected to a different controller?
CA1281521
Occasional Advisor

Re: MSA1000, Zoning and Redundant Switches

Dean,

Great, I will give it a try. Also the switch is the insternal one that the MSA1000 can have.
Uwe Zessin
Honored Contributor

Re: MSA1000, Zoning and Redundant Switches

No, it is because the controllers work active/standbay and only the active controller allows I/O to all logical disks. With Linux systems, it is possible that the boot of a single system causes a failover, so ALL servers must be attached to both controllers at the same time.

The internal MSA SAN switch 2/8 is based on Brocade.
.
CA1281521
Occasional Advisor

Re: MSA1000, Zoning and Redundant Switches

Sorry, but I don't see your point.

In case we have a single controller and two switches, I can connect 2 RAC nodes and 1 RHCS node to one switch, and the rest to the other switch. Here there are no failovers. If one switch fails, half the machines still can see the controller. Am I correct here?

When I add a second controller, given that I have no Multipathing software, no failovers can occur, so for me, my second controller is just a spare part sitting close to my MSA1000. Am I correct here?

For having a real redundant system, I need both Multipathing Software and dual HBA's to the MSA1000. Is MP software very expensive?

Thanks
Uwe Zessin
Honored Contributor

Re: MSA1000, Zoning and Redundant Switches

No, sorry, you are only partly correct. If the switch fails that is connected to the controller, then all your servers loose access to the data. If the other switch fails, then half of the servers can still access the one controller, but there is no failover (unless you have shifted the topic and you are now talking about failover on the application side).

When you add a second controller, it depends how your SAN looks like whether you see redundant paths or not. Almost no operating system has multipathing embedded and they can easily get confused when they see 'ghost' devices.

For a cheap solution without any management features you can use the embedded multipath feature of the Qlogic device driver from HP's web pages.
.
CA1281521
Occasional Advisor

Re: MSA1000, Zoning and Redundant Switches

From what I am reading, each switch is connectected to one controller only. Is it?
Uwe Zessin
Honored Contributor

Re: MSA1000, Zoning and Redundant Switches

If you put an embedded SAN switch into the MSA1000 box, yes, the controller directly connects to the switch at the back. Excuse me, but what is your point?
.