Installing T4 on cluster

[TMcB]

hi everyone

Just looking for some advice on installing T4 on a 6 node VMS cluster.
Just wanted to confirm if the following is correct (or not):
- install T4 on each node

- define t4$data to point to a separate folder for each node eg clu$disk:[t4.server1.t4$data] and clu$disk:[t4.server2.t4$data]

- create a batch queue on each node eg server1_t4$batch and server2_t4$batch

Thanks for your advice

[P Muralidhar Kini]

Hi TMcB,

>> Just looking for some advice on installing T4 on a 6 node VMS cluster.
The T4 kit comes along with a t4_readme.txt file. This has the steps to be
followed for T4 installation.
Have you checked the t4_readme.txt file, whether it already answers your question.

Regards,
Murali

[TMcB]

OK - I see that now, thanks

[Volker Halle]

TMcB,

you need to install T4 once on EVERY system disk in the cluster.

You can/should use a common T4$DATA directory.

You need to USE node-specific T4 BATCH queues, but you can use existing node-specific queues.

Volker.

[TMcB]

I split the data directories into separate subfolders :

eg clu4disk:[t4.server1.t4$data] and clu4disk:[t4.server2.t4$data].

Why would you have to lump all the data files together - what benefit do you get doing this.

thanks so much

[Hoff · Accepted solution]

Am I missing something?

It's easier to have one data directory because then you don't have to go chase around to obtain the data from various servers; you can roll up your results and look across your servers from one spot.

Given this is a cluster, so a server that's busy honking on a shared resource can derail performance on other servers. (Same applies for a hypothetical SAN controller overload case, as the potential for disparate server architectures and operating systems can make that case harder to inspect.)

And you're right, T4 should deal with this cluster stuff rather better, but then T4 has various (other) issues, too.

[John Gillings]

T,

T4 gathers data per node, in a node specific file, but sometimes it's interesting to compare statistics across nodes. To do that you either have to open the corresponding files from each node, or to use a utility like CSVPNG to merge files. In either case it's easier to do that if they're all in one place.

I keep all T4 data in a cluster wide archive. It's organised as a directory tree with one directory per day. I have a procedure which uses CSVPNG to select a time slice. It gathers all the relevant files, merges them, then extracts the time period of interest.