Sharing Jamaica drives - best way to terminate SCSI?

[Joseph Crawford]

I have two K-class servers, both attached to a Jamaica disk enclosure. The left and right disks are connected via a SCSI cable. If I want both servers to see all disks at the same time, do I have to buy self-terminating cables or is there a way to provide termination on the SCSI cards

P.S. This thread has been moevd from General to Disk Array. - Hp Forum moderator

[Angus Crome · Accepted solution]

First off, although, you can set them up to see the drives at the same time, you can only write from one machine. The other has to be mounted read only. Typically, you would see this configuration in a lab teaching Service Guard concepts.

How to do it;

First, you will have to change the SCSI ID of one of the cards from 7 to some other number (usually 0 or 15). Don't have a clue where you change it, but it is probably set on the card somewhere.
Next you will either need to connect on card to one each of the free connectors on the Jamaica. This is the easy way, but doesn't lend itself to fault tolerance very well.

A better way is to disable termination on the cards and use a couple of Y-SCSI cables to connect it the same way. This gives you the ability to shutdown or remove one of the servers without disrupting the other one.

A note, the Cards have internal termination on be default, since in normal configurations, they will be a terminal point for the cable.

All that said, there is not a lot of good reason to do this unless you are setting up a Service Guard environment or are setting up Oracle Parallel Server.

It may even be cheap enough to get a small NAS device and connect using CIFS or NFS over 100Base-T or 1000Base-T ethernet.

[Paula J Frazer-Campbell]

Joseph

Have a look at NFS and CIFS.

Paula

[Joseph Crawford]

Angus, thanks for the response. Actually, you are right on - I am setting up a ServiceGuard environment. I'll open one of the boxes and try changing the SCSI ID on the card.

One thing concerns me. You mentioned that if I proceed with this configuration I would not be able to shutdown one box without affecting the other? Am I understanding this correctly? If so, this would defeat the purpose of my ServiceGuard solution, right? Thanks.

[Angus Crome]

Unless you can terminate outside the SCSI cards, shutting off one of the boxes would remove termination from one end of the chain.

I would really recommend the Y-SCSI cables, but you will have to contact HP to find out where to get them. They had them in my service guard class on all D380's.

[David Bell_1]

Joseph,

In the MC/SG environment, each application is set up as a "package". That said, if one server is required to shutdown, you simply switch the package over to the other server. This will require the application be off line for as long as it takes to come up on the other server. This is not typically a long time (depends on the application).

As to the cables, you'll need the 'Y' cables as well as an A3591A (I think that's the number). The A3591A is an "inline" terminator. This is to prevent SCSI interrupts, etc. on a shared disk. MC/SG will prevent the shared disk from both hosts tripping over each other.

As for the SCSI cards, I'm assuming you're using either the HP-PB A28696A Full Height FWD SCSI or the A2969A HP-HSC FWD SCSI. Both of the cards utilize dip switches on the SCSI card itself to set the SCSI ID. I would recommend that you set ALL scsi ID's on one host to the default of 7 and ALL scsi ID's on the other host to 6. This can prevent accidental corruption if a cable is moved, etc.

Whether or not both hosts are using the disk at the same time depends on your configuration of MC/SG. Whether you choose an "active/active" configuration or an "active/passive" configuration. Active/Active means that both hosts will be running some application and if a failure occurs, the package(s) will fail to the other host. Active/Passive means that one host will run the applications and the other will be "waiting" for a failure to occur to take over.

HTH,

Dave

[David Bell_1]

Joseph,

My mistake, the product number which appears to be a legacy product number is C2980A for the M/F HVD SCSI ILT (In line termination)cable. Please see the pdf at:

http://www.hp.com/racksolutions/prodinfo/datacables/ds_scsi-legacy-cables.pdf

You can also see:

http://www.hp.com/racksolutions/prodinfo/datacables/ds_scsi-inline-cables.pdf

The verbage associated to MC/SG from the Configuration guide is as follows:

In-line terminated cables are used in MC/ServiceGuard and High Availability MC-certified systems. They eenable the host bus adapter to be serviced without interrupting the accessibility of the SCSI bus by other hosts. If the host bus adapter card has auto termination, it needs to be defeated by adjusting jumper shunts per the host bus adapter User Guide instructions. In-line terminated cables allow servers and host bus adapter cards to be serviced without inhibiting the remainder of the SCSI bus by removing termination dependency from the host bus adapter.

HTH,

Dave

[Brian M Rawlings]

One point of clarification:

If you DON'T have ILT (in-line terminated) SCSI cables, you don't lose SCSI connectivity in the shared bus if you one server. The termination (resistor packs) on the card continue to work (term power is provided via the SCSI cable, while other devices are live on the bus).

In other words, you can lose a server power supply, and MC/SG will still be able to access the data over the SCSI bus from the shared storage device. It will mount all necessary volumes & file systems up on the other server, and restart your apps according to the failover rules you have set up.

HP comes out, replaces the bad power supply, and the server comes back up. Data and Apps have been available and working the whole time it was down, from the other server. You manually "fail back" apps where they should be running, and life is good.

The reason for ILT cables is in case the failure is a SCSI HBA. Once again, either I/O fails over to another link (PVLINKS), or the apps fail over to the secondary node. Now, HP comes out to replace the failed SCSI HBA, and... Oh no, we can't pull the SCSI cable off the card, that would leave this end of the bus UNTERMINATED, and no data could flow over it until we got the new card installed and the cable plugged back in.

This is where the ILT cables come in. With termination disabled in the card, and terminators embedded into the plug ends, when you unplug the cable, the bus termination remains with the bus, and data continues to flow. So, the failed HBA can be replaced without interruption of the I/O on that channel (the I/O to the secondary node). Again, life is good.

Since ILT cables are really about "continuing to operate through SCSI HBA failure & replacement", with no downtime to make the repair, they really fall under the category of "must have for production", not "must have". It would be silly to worry about this high-uptime component in a lab or classroom, other than to teach about it and demonstrate why you'd want it in production.

A couple of minor but related items:
1> SCSI rules state that "termination must be present within .5 meters of each end of the SCSI bus". This is true for all types of SCSI (SCSI-1, SCSI-II, UltraSCSI, LVD, HVD, SE, all of 'em).
2> The only time the three-connector "V-cables" (or "Y-cables, same thing) are needed is if you are doing a 3- or 4-node cluster with SCSI shared storage devices. For 2-node clusters (the vast majority), you connect one server (via normal cables) to one port of the shared storage device, and the other server (via normal cables) to the second port on the shared storage device. You don't need more than two ends on the whole bus, with a server on each end. Unusual cables (V- and/or Y-cables) are only needed if you need more connectors on your SCSI bus, to connect additional hosts/nodes beyond the original two.

Sorry to add to an old thread, but I felt the clarification was needed for future researchers.

Regards, --bmr