Runing batch simulations across many unix machines.

richardbradley

The best aspect of starting at a new company is seeing all the cool little tools that were written in-house to make our lives as verification engineers simpler. There are a lot of clever scripts and programs out there that are actually quite useful.

Here is a much improved rewrite of a tool I wrote several years ago to batch simulations across multiple *nix machines. I know several groups still using the old version. Hopefully others will find it useful also.

I have tested it as well as I can with the limited resources available to me at the moment (mainly a couple Linux boxes) and had a couple people use it in a real project, but it wouldn’t surprise me if there were more bugs lurking in the code. Any feedback is welcome and encouraged. Actually, just let me know if you find this sort of thing useful.

I am releasing it under the Artistic License. Just type "parallel --license" to see it.

Code tarball is here:
http://richardbradley.org/hdl/asic_verification.html

From the documentation intro:

Parallel is an attempt to give a verification engineer the ability to batch simulations across dozens of 'nix machines. It uses no daemon of its own, but instead relies on NFS (lock files), and either ssh or rsh to communicate with remote machines. Hopefully these tools are already installed and configured in your system.

It is a no-frills (At least low frills) solution. There is some load-balancing abilities, but parallel is not very good at it. There is no attempt to integrate with any given simulator. No attempt is made to make it easier to suspend and then restart a simulation, but there is nothing to stop you from adding that yourself.

What it does give you is a simple way to setup, archive, and run a series of regression tests across an arbitrary number of machines and report the status of them as they run. Its non-central design allows for many designers to have their own setup of their own tests with little overlap with other groups, or designers. The fact that it uses no daemon of its own makes it easier to setup and less likely that you will need to call the sys-admin at 3:00 in the morning. And best of all it is (hopefully) written in well documented Perl code, so you can do whatever you want to it, and openly make fun of its author's poor coding and spelling abilities.

Enjoy,
Rich