OnDemand User Group
Support Forums => MP Server => Topic started by: Marc on November 29, 2011, 02:17:58 PM
-
Does anyone use an active/active configuration, i.e., two physical servers, each running their own CMOD, but sharing the same DB2/TSM storage?
If so, what technologies are you using to handle contention within DB2 ? Contention can be an issue because when a document is retrieved, it causes a write operation to log the activity in the System Log, within DB2. Simultaneous writes to DB2 can cause contention.
We currently use CMOD 8.4.1.4 and DB2 9.5 but plan on moving to CMOD 8.5 and DB2 9.7.
-
Hello Marc,
don't do it, an active/passive is ok, and will be always ok.
But an active/active is, at my knowledge, not supported by CMOD.
Maybe others have other experiences...
Sincerely yours,
Alessandro
-
I can't imagine a design where you have two active CMOD servers sharing a single DB2 database. That eliminates any benefit you have to running redundant hardware -- which is usually the least likely piece of the architecture to fail, really.
Most active-active architectures are entirely separate. Different instances, running different databases, storing on different disks, hosted on separate servers, living in separate data centers.
How are you tying these two servers together?
-JD.
-
We are currently active/passive on our primary site (two physical servers, each running their own CMOD, but one shared DB2), AND we have redundant hardware on our DR site (also passive, with separate physical server and a second DB2).
We wanted to investigate active/active on our primary site to:
1) reduce the outage during a deployment
2) make use of hardware that is sitting idle (passive)
3) increase throughput of online retrievals
To Justin's point: We have the redundant hardware on DR site.
How would Active/Active with different databases be achieved? CMOD doesn't support replication from one database to another, so to keep the data in sync, I think you would have to load all the data twice, once to each database.
-
We have active/active with each instance having its own CMOD, DB2 and TSM. The servers are located in different data centers in different cities. We have a frontend process on a LINUX based server that does load balancing and sends each input file to both instances thus allowing the files to be loaded concurrently. We switch between servers as maintenance/upgrade is required on each server and our web services switch to the hot active instance as required.
Russell
-
I recommend the configuration that Russell is using.
-
Hi,
on Z/OS another combination might work if you have a sysplex and a db2 sharing group.
1. define a virtual IP
2. start ARSSOCKD on both images of the sysplex listening to the virtual IP
3. If ARSSOCKD has implemented a proper rollback / commit logic it shall work
With this technique you would have a simple workload and the system will still work if one image is down. I think it will work, because we are running several ARSLOAD in parallel with direct DB2. This is similar to more than one ARSSOCKD running against the same DB2. But of course IBM should state that it works.
regards
Egon
-
For the active/active scenario with each instance having its own CMOD, DB2 and TSM, I don't think Annotations would work.
Annotations are tied to a specific object, so if you were to add an annotation, it would only be in one of the databases. You can't control which of the two objects you retrieve next time, so you may or may not get the Annotation.
-
It's possible to replicate Annotations across sites.
-
Justin, can you elaborate on how to replicate annotations?
IBM has said that CMOD does not support replication, so I know it's not done directly by CMOD.
Also, I don't think DB2 HADR can send data both ways.
-
It's been a while since I set this up, but it basically involves setting a series of triggers that find annotations, the record on the opposite machine, and duplicates the annotation.
-
Hello - reviving an old discussion here. I'm looking for a solution to sync annotations between two data centers. We have a current home-grown solution but we're looking for a better way.
Our setup is active/passive for users (who use annotations). Active/active for internet traffic (no annotations). Our systems load concurrently to both data centers. There is no replication with our current setup.
Thanks everyone!
-
As of today, it is not possible. You must use you current setup.
It is possible that this active-active scenario will come to CMOD is a short-medium term future, but until it arrives, you will need to live with the current solution limitation :-/
-
Hi Marc,
just an additional remark because of new experiences.
We are now using two arssockd and 10 arsload instances on the same machine against the same CMOD database. We don't see any retention problems. At least on Z/OS machines it works fine.
regards
Egon
-
Hi Marc,
just an additional remark because of new experiences.
We are now using two arssockd and 10 arsload instances on the same machine against the same CMOD database. We don't see any retention problems. At least on Z/OS machines it works fine.
regards
Egon
Well that's something that works ONLY on z/OS today... but not yet on MP.
I have heard that they will have it on MP in the future. When in the future I have no idea... but it is coming. I hope not in 10 years.... :-/
So an answer would be, you want active/active with CMOD? Then switch to z/OS !!! :-D
-
We are using a Active/Standby HA utilizing HPE ServiceGuard.
We have two identical (CPU, RAM, etc.) servers in a cluster at one site.
All storage (cache, db2, tsm, application) is stored on SAN disks that float
between the nodes of the cluster.
If there's a hardware failure on the active node, it takes less than 10 minutes to failover to the standby node.
-
We are using a Active/Standby HA utilizing HPE ServiceGuard.
We have two identical (CPU, RAM, etc.) servers in a cluster at one site.
All storage (cache, db2, tsm, application) is stored on SAN disks that float
between the nodes of the cluster.
If there's a hardware failure on the active node, it takes less than 10 minutes to failover to the standby node.
Nice, setup :-)
but the original question was active/active .
-
Active/Active is only available on z/OS.
Ed
-
Active/Active is only available on z/OS.
Ed
No longer true as of CMOD v10.1.
High availability active-active is now available on AIX.
Ed
-
On CMOD 10.1, support Active/Active architecture as this technote:
http://www-01.ibm.com/support/docview.wss?uid=swg27049639