Web Server:
a) Current usage and statistics of cluster
b) Web Service for registering password c) Internal Web Page for user list - for Grid FTP
2. DU Server as Head Node:
a) Head node and compute nodes b) License Server for MATLAB compute nodes
c) Scripts for copying data to SQL Server about user jobs
3. SQL Server
a) Lookup Table for users (used from Web Server and Head node)
b) Data stored from DU server.
4. Proxy Server
5. Grid FTP Server - For moving data, used in simulations
6. DataDirect Network Storage -8 TB (Approx) of dedicated storage for "MATLAB on the TeraGrid" Could be increased or decreased after getting usage and user data.
User and Cluster Model:
User Desktop or the Client Machine will store matlab jobs on the local drive
…show more content…
Each worker can access the shared drive on the network to look at the job configuration.
After the workers have finished computing the client can access the output from the shared drive.
Scheduler Architecture:
There are independent queues for managing the job efficiently, when the job is submitted initially it moves to the pending queue, from the pending queue it is submitted to the queued queue where it stays for some time until the server can begin to compute it. Then when the server is ready to send tasks to the workers it can or has already created it accepts the job keeps it in the running queue where each job is submitted to a worker.
Further proposed optimizations on the Model:
Multiple schedulers can be incorporated at the server that can spin up their own individual Virtual Machines as their workers. Each Client can submit jobs to one or multiple schedulers. Each scheduler after getting the output can push it back to the client that requested it. Individual workers are created on the virtual machine created by the