Page 1 of 1
how to know SMP or MPP or CLUSTER or GRID
Posted: Mon Feb 27, 2012 3:45 am
by karthi_gana
All,
is there a way to identify the server (without asking the ADMIN team) is SMP/MPP/CLUSTER/GRID?
i can see 3 nodes in the director log.
Code: Select all
main_program: APT configuration file: /opt/IBM/InformationServer/Server/Configurations/default.apt
{
node "node1"
{
fastname "xxxx.com"
pools ""
resource disk "/node1/res" {pools ""}
resource scratchdisk "/node1/scratch" {pools ""}
}
node "node2"
{
fastname "xxxx.com"
pools ""
resource disk "/node2/res" {pools ""}
resource scratchdisk "/node2/scratch" {pools ""}
}
node "node3"
{
fastname "xxxx.com"
pools ""
resource disk "/node3/res" {pools ""}
resource scratchdisk "/node3/scratch" {pools ""}
}
}
can i get the required information from this?
Posted: Mon Feb 27, 2012 4:26 am
by pandeesh
If it is Grid, you can easily identify in the log. There will be a script called "DynamicGrid.sh" which will dynamically allocate the resources.
Posted: Mon Feb 27, 2012 8:43 am
by priyadarshikunal
also the fast name is the server hostname if there are different fastnames that means there are multiple servers involved. Also there will be a conductor node in case of Cluster/Grid.
Posted: Wed Feb 29, 2012 7:59 am
by karthi_gana
yes. The "fastnames" are different only. so it means there are three server involved.
On what basis nodes are added ? i mean how should i decide the node count to add it in a particula servar? Say for example, I have one node for each server. if i want to add two node for a server, can we just add it? or do we need to consider some parameters before adding new nodes?
Posted: Wed Feb 29, 2012 8:33 am
by PaulVL
Also the "default.apt" indicates that it was not dynamically generated, so that rules out GRID.
Why do you not want to ask your admin team?
As an admin I WANT / BEG new developpers to introduce themselves and ask questions on the environment. I'd much rather see someone come to me with a question than someone go down the wrong path because they did not.
Posted: Wed Feb 29, 2012 2:57 pm
by ray.wurlod
Every node you add increases spot processing load (number of processes). That's the main "parameter" you need to consider.
Too many nodes can be counter productive. For a silly example, eight nodes to process one row equals seven idle nodes, but the processes all still need to be started. For a less silly example, if you partition on gender all but two nodes will be idle (except, perhaps, in California).
Posted: Fri Mar 23, 2012 3:41 am
by karthi_gana
what kind of actitivities will be stored in resource disk? first of all, what is mean by resource disk? It is is a kind of Hard Disk. Am i corect?
As for as i know, the lookup data will be stored in scratch disk. Right?
It is a kind of RAM? Am i correct?
Posted: Fri Mar 23, 2012 10:13 am
by qt_ky
Both resource types point to paths, so yes they are on disk, not in memory.
Please consult the Parallel Job Developer's Guide chapter "The Parallel engine configuration file" for details.