05-21-2014 10:30 AM
I am new to SAS and I have been asked to install v9.3 on our HPC linux cluster. I wondered if anyone had any documentation for planning and installing SAS 9.3 on a HPC cluster. I'm very much at sea. I've taken a stab at working throught the installation, however there were a number of points that I did not fully understand. The main issues are:
Our license appears to suggest that this is a "planned deployment" and therefore a deployment plan is needed. That does make sensem however how to I find/access the plan?
Should I create a system wide account called "sas" and then us this for the installation? Trivial point perhaps. I started to run setup.sh as root, however should I work as "sas" from the beginning.
The most confusing issue was that the installation script/wizard appeared to want to lock my installation to the machine that I was working on. This happened to be one of the login nodes, however I do wish SAS to be visible/usable accross the cluster -- on both login and compute nodes. Am I correct in thinking that the installation will comprise of server and client elements -- do you actually run a server/service on one of the nodes in the cluster and then use the client infrastruction to communicate with the server processes?
Apologies for appearing so green. Your advice would be appreciated, please. As I say it might be best if someone has a basic installation guide for SAS on a computing cluster.
Best regards -- David.
05-21-2014 11:21 AM
Deployment plan: usually, you can chose from standard plans in the configuration wizard. If you have problem matching the plan to what you have licensed, or you want a site specific plan file, contact your SAS reprasentative for guidance.
User account: don't use root. Most Unix (and Linux?) admins wants to keep the use of root to a minimum, for security reasons. SAS recommends quite clear that a specialized install account should be used, 'sas'. So yes, use sas for installing.
Cluster: I'm not familiar with HP Clusters particularly. Try to describe how you wish to use this cluster with SAS (i.e. which SAS products you have licensed), and how the cluster is configured.
05-21-2014 11:46 AM
Thank you for your reply -- your points were very helpful. Perhaps I shouldn't use acronym's like HPC, however I'll clarify that point now. I'm referring to a High Performance Computing cluster. Essentially we have a large number of Linux machines networked together, and all the machines see common file stores providing the users' directories and the applications (including SAS of course). Three of the machines serve as login nodes, and the rest as compute nodes. There are also special machines to manage the file stores. Users submit jobs from the login nodes and then a scheduler and manager (moab and pbs/torque in our case) distribute the workload across the compute nodes. This type of setup is not uncommon. Users will submit jobs to use SAS to analyse their data on the cluster.
That's the basic setup. As I say, my confusion arose from a couple of things. First of all there seemed to be 3 software cateogories to install -- server, client and mid tier. Having only worked with STATA I was confused by the complexity of the SAS set up. Also the installation seemed to suggest that the installation was perhaps bound to the machine I was working on, however that may not be true in reality. I am. afterall, making the installation on the shared application file store.
Best regards -- David.