12-11-2014 12:38 AM
For Example I have two instance (lev'X' and Lev'Y') and I need to do workspace load balancing for this two instance.
Question 1:Can we do Workspace Balancing between two workspace server on the same machine but different instance and server running on different port.?
Question 2: What are component need to install for the Workspace Loadbalancing on both the Lev'X' - Lev'Y' for configuring Workspace Load Balancing.?
Question 3: Would the metadata server be common or separate for the two workspace server.
Question 4: Would object spawner be different or same for the respective workspace server and can we have two object spawner on the same machine with different port contributing to workspace load balancing.
12-11-2014 08:03 AM
First, you don't use a Grid?
1) what's the point of balance the load on the same server? Or do you mean the Pooled WS server?
3) To do laod balancing, the balancing server needs all information on the servers that will take part. Any other called servers are only self aware.
Given the complexity in your questions, my advice is to contact SAS or associated partner that can help you out with the architecture and design.
12-11-2014 01:52 AM
Workspace server are instances (Operating System) processes used by EGuide DI AMO etc to do some work at the Operating System level.
SAS is not controlling the Operating System that is RedHat (Linux) Microsoft (Windows) IBM(Mainframe).
Do you want to do Workload balancing you are with SAS at wrongly addressed.
Sorry for this message, better be clear on that.
What you can do with SAS is:
- using LSF (coming from IBM) for load balancing in a grid
- using cgroups at the Linux environment as coming from RedHat.
- some balancing for the shared services (Stored Process , Pooled workspace) using shared accounts for special logical functions.
You have to evaluate whether these logical functions and the shared usage are acceptable and applicable. (business requirements)
These are not meant for controlling the limited resources on a system but for decreasing response times as the start initialization is bypassed (shared).
This is the loadbalancing reference Linus has included.
Think see many requests coming from the internet and you need more workers to be quick enough to handle those.
This is under the assumption your system itself is fast enough to handle al those request. (no grid) It is extensible to use some grid approach.
- The Lev-x Lev-y is excellent for DTAP version and release management of the business artifacts. There is NO relation to loadbalancing.
As metadataservers are OS processes you could control them in a clustered environment with cgroups for the DTAP.
You will need using different service account for that.
- The objectspawner is the onde doing the forks (runas sudo surrogate) in a SAS manner.
It makes not lot sense to have multiple ones of that. It is not a very resource intensive task. As it does the forks it is the template for all processes being started. At Unix/Linux be prepared to see all properties of this user is going to all the childs and these childs are not set as expected by security people. It is the sasauth saselsrv at rootlevel wiht setuid playing a role in this. At windows it is running at the local service account for this reason.
- There is no design in the metadata for managing servers at that low level. Metadata is a logical concept and the SASapp servers are logical context for buinsess environments representing a business process (application) or a business organization (hierarchy).
Any questions on this?