WP_Term Object
(
    [term_id] => 119
    [name] => Runtime Design Automation
    [slug] => runtime-design-automation
    [term_group] => 0
    [term_taxonomy_id] => 119
    [taxonomy] => category
    [description] => 
    [parent] => 14433
    [count] => 15
    [filter] => raw
    [cat_ID] => 119
    [category_count] => 15
    [category_description] => 
    [cat_name] => Runtime Design Automation
    [category_nicename] => runtime-design-automation
    [category_parent] => 14433
    [is_post] => 1
)

Farm Management

Farm Management
by Paul McLellan on 03-01-2012 at 5:34 pm

Every so often I come across a new company in EDA or one of its neighboring domains, new to me anyway, and new to SemiWiki. One such company is RunTime Design Automation (RTDA). They provide a suite of tools for managing server farms (or internal clouds which seems to be the trendy buzzword du jour). Running a few EDA scripts on a few servers is something that is not too hard to do, but when you are looking at running tens of thousands of jobs on thousands of servers, you have a whole new set of problems. Servers crash or hang. Tools run out of licenses. Jobs depend on each other in a complicated tree. These are the problems that RTDA addresses.

When I was at Ambit we had a Q/A farm of 40 Sun workstations and 20 HP workstations, an unusually large investment at the time. Now this is a trivially small farm in an era when data-centers may house 100,000 servers and literally millions of jobs a day need to be scheduled.

RTDA have four products:

  • LicenseMonitor
  • NetworkComputer
  • FlowTracer
  • WorkloadAnalyzer


The simplest product is LicenseMonitor. It pulls license usage data from the license servers used by EDA (and other) tools, such as FLEXlm. This gives a summary of what license usage really is so that a company does not end up either having too few licenses and thus wasting time when running jobs, or having too many licenses and wasting money. It also interfaces with Network Computer to ensure jobs are not scheduled until licenses are available.


NetworkComputer is the highest performance job scheduler to spread a workload over a large server farm. It is similar to the well-known LSF but more powerful and better integrated with EDA license management. It has a GUI that allows a user to keep track of thousands of jobs, which are color coded as they become runnable, start to run, crash, complete and so forth. It can handle millions of jobs on thousands of processors. It can get license usage of, for example, simulation licenses up to very close to 100%.


FlowTracer is a much more sophisticated tool like the Unix make command that manages complex design flows, handing all the dependencies, errors and so on. It handles the entire design flow giving visibility to what is happening and automatically updating dependencies intelligently.

Finally, WorkloadAnalyzer is a server farm simulation allowing efficient compute farm planning, answering questions such as what would happen if additional servers or additional licenses were purchased. It can be used daily for planning or annually to provide data for license renegotiation. It can take information from Network Computer (or its competitors) and analyze how that workload would run under different circumstances and thus make it easier to optimize the investment.

Download the free e-book, The Art of Flows here


Comments

There are no comments yet.

You must register or log in to view/post comments.