Any better solution than MC/SG?

[DeadUserIDCA1084750]

Hi, all.

Our shop is running on Service Guard with HP-UX 11.0 OS. We had it for 2 years and were quite disappointed with it. Firstly, our application(SAP) needs to be down during the failover. Secondly, it requires a lot of human intervention during a failover.

We are wondering if HP has a solution that can do a more "seamless" failover. That means, the application is not affected during a failover. Probably hardware cluster?

Please advise.

[Ashwani Kashyap]

Yes , during the failover the application will be unavailable , that how MCSG works .

FOr seamless solution , I don't know , may be NON STOP OS clustering solution on Tandem .

[Mel Burslan]

what you are looking for is called a non-stop server configuration. I can not recall the name of the 3rd part providing this solution but it is not straight hpux. It is actually known by a different OS name but underneath this OS, there lies multiple hpux run boxes running in parallel. Of course, it has some additional commands associated with it.
One downside is : they are prohibitively expensive. The only people that I know using these boxes are big, very big brokerage houses on the NYSE floor where stopping is not an option.
Also there is another solution from HP called himalaya but they are actually Compaq branded Wintel boxes, which I am not sure suitable for running large SAP instances.

[Stephen Doud]

A failover of a Serviceguard package, whether SAP or not, should not require attention unless the failure includes dead disks/PV-links. If you are not certain that your cluster and package are configured correctly, seek the aid of an the Remote Solution Center or an HP consultant to review the current configuration and make improvement recommendations.

The fault-tolerant system known as Non-Stop (which HP acquired with it's merger with Compaq) is described in this site:
http://h71033.www7.hp.com/

-SD-

[Geoff Wild]

What you are asking is more based on the application - that is SAP - needs to be made more cluster aware....there are more and more features coming to MC/SG - some from Tru64 soon....You can do some better HA with Oracle Parallel Server....

Rgds...Geoff

[Mark Grant]

Actually, IBM's Highly available solution can do this if the application can do reverse arp caching. The same is probably true of service guard. Although I tend to think that doing it manually or even scripting it is not that inferior to service guard (and much cheaper) it must be said that service guard is very configurable.

The point is that the problem is the application, not service guard as such. If the application is written to cope with a few minutes of itself and it's disks going off into a big black hole then it would be seamless. You might even be able to get away with being a bit creative with your control scripts to allow even non service guard aware applications to play nice during failover.

Also, if a system has failed, do you want the application to run seamlessly on another machine when there could be something seriouly wrong with the application itself? Personally I disable automatic fail-over where I can. I want to have a quick look before starting things up again.

[Steven E. Protter]

Veritas and Oracle both have solutions that provide some of ServiceGuard's functionality.

I just took a SG class and kind of like it.

I will say this: My instructor was a consultant to the team thats building the next SG. Its going to be based on True64 Cluster and apparently bare little ressemblance to SG. So says my instructor, I'm just acting as a reporter here.

So there is hope HP will produce something you like better, if you are patient.

SEP

[Rita C Workman]

Well Jonathan, folks have mentioned a couple products for 'seamless failover'.

But I've been using MC/SG for a little while now, and if you have to have 'alot' of human intervention, then I'd say your MC/SG is configured dead wrong. The only time I have to intervene is to decide if an entire site has failed and do I really want to initiate a full SITE failover. Everything else is automated..

So maybe before looking to another product, why not see if you can get the SG you have working the way it was intended.
And yes, it will stop the database on one system when it fails....but done correctly it will bring it back up on the other, all automated.

Just my 2cents,
Rita

[Bharat Katkar]

Hi there,
Veritas Cluster would give you probably more better results which come along with more complex structure but with added functionalities.
Regards,

[DeadUserIDCA1084750]

Hi, all.

Thanks for the feedback. The primary concern I have is that SAP goes down with the server. I've got some experience with Oracle RAC. One server goes down, oracle doesn't and data loss is minimized. That is what I expect from MC/SG. Geoff has a point that
SAP should be more failover friendly or Oracle RAC friendly. But if SAP can handle failover, MC/SG has little value.

Sorry that I use SAP as my example because that is the main application MC/SG is used for. Yes, the SAP package failover is automatic. But depending on the type of failure, you need to do extra tasks like taking out the failing node from the cluster, packages, etc after the failover happened.

Now when you think back, the MC/SG SAP packages are just some shell scripts that remount file systems from your failed server and start SAP again. These scripts are not hard to write on your own. Also, when you modify your SAP environment like adding additional mount points(sapdata), you need to modify the MC/SG script anyways. That means we need to run production tests.

Now it comes back to my original question: Is there another better solution out there on the market that runs with HP-UX servers? What does MC/SG bring to the table?

[G. Vrijhoeven]

Hi,

I agree about the down time during fail over and if the availability is provided by RAC the need for MC/SG is less.
But i do not agree on the easy to script part:
a few advantages:
MC/SG has standby lan interfaces that work fine.
The volumegroups exclusive mount option.
The monitoring scripts ( written yourself, but operated by MC/SG)
Backup plugins.
Split brain.

But may be i underestimate your scripting capabilities.

HTH,

Gideon

[Mohanasundaram_1]

Hi Jonathan,

Interestingly you spoke about Oracle RAC. There is a SG extension for RAC, which may help. But I am not sure about SAP.

Service Guard is to provide a hardware redundancy to your environment like SPU failure, LAN card failure, HBA failure etc.

The database+application should handle such failure to provide you a seamless performance. The Oracle RAC does that at the DB level.

As others mentioned, you have the hardware available for the purpose with a cost.

Also, fail-over does not require human intervention until it is designed so.

are you trying to do away with serviceguard?

Cheers,
Mohan.

[Ted Buis]

HP offers ServiceGuard Extensions for SAP R/3, part number B7885BA (HP Somersault). This extensions for SAP also needs ServiceGuard NFS toolkit, B5140BA. Have you investigated these offerings? Also, HP Cluster Consistency Monitor (H8395AT) can help maintain a SG environment. Does your environment allow for a transaction monitor product, like BEA Tuxedo?