Sneak Peek at CPLEX 12.0

I am at the INFORMS Western Regional Conference. It is in Tempe, AZ and is right before the Practice Conference. So, we got a sneak peek at some of the new optimization software before tomorrow’s pre-practice conference technology workshops where Gurobi, Xpress-MP, AIMMS, LINDO, Microsoft (solver foundation), Frontline (makers of Risk Solver platform), Maximal Software (MPL), SAS as well as CPLEX and others will all have workshops. It is an exciting time for the software side of operations research and I am sure you’ll soon be able to find more information about these workshops at the conference’s official blog. But here’s a sneak peek at CPLEX 12.0:

{Before I continue, I just want to mention that the information below comes from the notes I took and not from official sources so I hope there is no error.}

I think one of the most important changes in CPLEX 12.0 is its interoperability with other software. This is great news both for practitioners and researchers in operations research. What I mean by interoperability?  In this version of CPLEX,

• you’ll be able to connect with MS Excel (both through macros and directly from a user interface),
• you’ll be able to use it with MATLAB, and
• it allows Python plug-ins or more correctly Python modules.

I don’t know much about Python but you can find more information at William Hart’s blog. I met Bill at the conference and as he was explaining, this is like Firefox’s add-ins that allow one to increase the functionality of the software and allow for customization. This, I think, might be a great thing to have.

On the technical side, some of the new developments in CPLEX 12.0 include:

• multicommodity flow cuts, and
• enhanced heuristics among other things.

For parallel processing,

• the shared-memory parallel processing now comes under the standard licensing.

Please note that this is only for shared-memory parallel processing (multi-core machines) and not for distributed-memory, clusters of computers, etc. Barrier method gets on average 1.4-2 times performance speedup and MIPs get 1.3-1.75 times on 4 threads.

One other thing to mention: The effectiveness of CPLEX 12.0 becomes more apparent on ‘hard’ problems. For instance, initial tests indicate for models that solve within 1 second (985 models total), 90% of them tie between CPLEX 11 and 12 and 7% of the time CPLEX 12 wins. However, for problems that take between 1,000-10,000 seconds to solve (292 models) CPLEX 12 wins 61% of the time with 11% tied for both versions. Note that a win is defined if a version is faster than the other by at least 10%.

It is an exciting time for OR software.  The accomplishmets so far have been great but there is much to do and this is such a crucial step for OR to become more widespread, so, I hope progress comes fast.