We are a very fresh startup focusing on OR (mainly within fish/salmon and energy (power and oil&gas)).

Seeing that we are a startup we are not awash with cash and are during during the early startup phase looking to use free/open source solvers. However, given the benchmark results that I have seen between different solvers (especally for other than strictly linear problems) and that the three main commerical solvers (Cplex, Xpress and Gurobi) seem much faster, I wondered if anyone knows if these companies/solvers have any special deals for startups? (My google search on the topic did not produce any meaningful result).

The solver we would need would need to handle a multitude of problems, though initially we need a good solver for quadratic programming problem. Our previous experience with solving such problems is with using Xpress, and if none of these companies offer special deals for startups, we plan to use this (http://www.numerical.rl.ac.uk/people/nimg/qp/qp.html) as a starting point for finding a good solver.

Cheers, Erlend Torgnes www.optimeering.com

asked 11 Nov '13, 07:56

ErlendT's gravatar image

ErlendT
101114
accept rate: 0%

edited 11 Nov '13, 08:00

Thank you all for the promt answers, will be sure to pursue the suggestions. From Hans Mittelmann's bechmark test for QP's it seems like the open source solvers have a bit to go before they can match the commerical solvers (at least for solvers in his benchmark)

(13 Nov '13, 04:08) ErlendT
2

The benchmark only say something relevant if your problems somehow resembles the benchmark problem :-).

(13 Nov '13, 05:31) Erling_MOSEK

Is MOSEK the only vendor so friendly with startups?

(13 Nov '13, 13:24) Stefano Gual...

MOSEK is friendly towards everybody of course. So I suggest potential customers trys us out.

(14 Nov '13, 01:40) Erling_MOSEK
2

Open-source solvers fill a different need from commercial ones. They may not compete on raw performance, but you can't beat them for performance per dollar invested in license fees. If the performance is adequate to get your particular problem solved as fast as you need it, then they are worth considering. Also, if you need to get into the code to specialize it (beyond the hooks provided in the API), you need the source.

(16 Nov '13, 15:44) Matthew Salt... ♦

MOSEK solves convex QPs as well other optimization problems. We are very friendly towards startups. In fact we will normally let startups use MOSEK for free until they go live with their product and make money. Since we are not too large and lean paperwork is usually not needed in the begining.

I suggest you contact MOSEK sales if you are interested.

All the best with your project.

Erling (CEO of MOSEK).

link

answered 11 Nov '13, 11:10

Erling_MOSEK's gravatar image

Erling_MOSEK
61614
accept rate: 3%

You might want to take a loog at the COIN-OR project. We publish free open-source software tools for optimization. There are several MIP solvers and a variety of other solvers and tools (over 50 projects in all).

link

answered 11 Nov '13, 22:17

Matthew%20Saltzman's gravatar image

Matthew Salt... ♦
4.7k310
accept rate: 17%

I guess you could take a look at Wikipedia's optimization software list and just check each project's website or even contact them. In theory, you need to poc each one that falls into your budget, but that's a lot of work. Looking at results of independent research competitions (such as itc2007, roadef2012) might help to weed out some of the lower performance solvers (presuming that your use case is close enough to those competitions).

For my project specifically:

We have a number of startups using OptaPlanner (open source, Java, apache license) that are very active in the community (such as asking questions on the mailing list, talking on IRC, raising issues in JIRA and in some cases even contributing code). Those startups generally don't buy paid support (which is aimed at large companies that need mission critical support) and that's fine. And they get along fine with the free community support (mailing list, IRC).

Besides that, I believe Bo's project (Sulum) aims at providing a cheaper solver.

link

answered 11 Nov '13, 09:11

Geoffrey%20De%20Smet's gravatar image

Geoffrey De ... ♦
3.6k32764
accept rate: 6%

Thanks for the mentioning ! We don't have QP support (yet), so sulum is probably not a fit for OP.

(11 Nov '13, 09:35) Bo Jensen ♦

A very good resource on optimization software is the OR/MS Today Linear Programming Software Survey, available from here. Its table 9 provides information on which types of problems can each solver handle. In addition, you could search for cloud computing services of solver vendors, which allows you to solve your problems using their own computational facilities and bill you based on your usage (hence, no license is required). As far as I know only Gurobi offer this service, but perhaps there are other vendors active in this new market (see a previous discussion of this issue on OR-X here).

link

answered 11 Nov '13, 12:14

Ehsan's gravatar image

Ehsan ♦
4.8k31122
accept rate: 16%

Please contact me if you'd like to discuss how you could use CPLEX. We have flexible trial conditions and can discuss: j-f at fr dot ibm dot com CPLEX is used by quite a large number of independant software vendors, from very large ones to startups. You would not feel alone if you'd use it!

Regarding benchmarks I concur with what Erling says: the best benchmark is on your problem.

link

answered 16 Nov '13, 10:41

jfpuget's gravatar image

jfpuget
2.5k310
accept rate: 8%

You might want to take a look at Hans Mittelmann's benchmarks.

link

answered 11 Nov '13, 15:27

Paul%20Rubin's gravatar image

Paul Rubin ♦♦
14.6k412
accept rate: 19%

AIMMS can call CPLEX, Gurobi, MOSEK, and several other solvers that can solve QP's, as well as other types of math programs. Since you're on a start-up budget, AIMMS comes with IPOPT, an open-source QP/QCP/NLP/LP solver on The COIN-OR Project.

There's a fully functional, unrestricted licenses without any limitations on problem size to commercial users (30-day trial) and academics (forever).

Depending on your needs, AIMMS optimization models can be called through the GUI or as a component of another application. The models can be deployed as a local, stand-alone applications or as server-based application.

link

answered 14 Nov '13, 12:54

bobbyn's gravatar image

bobbyn
111
accept rate: 0%

Hi Erlend, would it be an alternative for you to develop a solver agnostic application or model, i.e. to have an open design where you can plug in several solvers? This might give you the opportunity to develop using open source software with e.g. Apache license and then to offer commercial solvers in case your client needs the additional performance (and you do not need to invest into solver licenses too early). There is also the question how you manage licensing: Does your customer buy the solver license himself (and takes care of updates etc.) or is that your business? This is a rather complex (but boring) topic around legal questions.

link

answered 08 Dec '13, 16:29

Andreas%20Cardeneo's gravatar image

Andreas Card...
1762
accept rate: 10%

What you suggest is likely the ideal solution and what we plan to do going forward by making use of Python and its framework for optimization as most (all?) of the commercial solvers and many of the open source ones can be called from Python

(09 Dec '13, 05:36) ErlendT

See live catalogues of the Decision Optimization tools at www.decision-tools.org

link

answered 12 Nov '13, 13:19

jacobfeldman's gravatar image

jacobfeldman
212
accept rate: 0%

I just checked www.decision-tools.org and there seem to be very little content. Note the NEOS guide has lot of content. See the LP section for instance.

(14 Nov '13, 01:49) Erling_MOSEK
4

Those Live Tool Catalogues have been created to help business application developers to choose the right decision tool for different tool categories. These catalogues allow a potential user not only to compare tools feature-by-feature in one large table, but also to see how these tools are supported, used, and how they are integrated with other tools.

The first catalogue was created and fulfilled by the start of the CP-2013 this September - see Catalogue of Constraint Programming Tools. It now contains 30 profiles of the currently supported CP solvers created and maintained by their authors. Only recently we started to send out invitations to authors of the tools s in other categories. In particular, the LP/MIP Tools Catalogue is a work in progress: it currently includes 13 tools and only 4 profiles have been completed by their authors. You will see more completed product profiles soon.

(16 Nov '13, 11:33) jacobfeldman

@jacobfeldman Great work by aggregating that data :) But why split it up into 2 distinct catalog's? From a user's perspective - both categories solve similar problems.

Having a "solver type" (CP, MIP/LP, MH) property per row is interesting, as long as all rows are shown in the same page.

Many users don't understand the difference between CP, MIP/LP and MH. And most experts (including myself) are biased against 1 technology (because they spend 80% or more time on that tech and 20% or less on the others), so they won't inform the user objectively.

(17 Nov '13, 04:46) Geoffrey De ... ♦
Your answer
toggle preview

Follow this question

By Email:

Once you sign in you will be able to subscribe for any updates here

By RSS:

Answers

Answers and Comments

Markdown Basics

  • *italic* or _italic_
  • **bold** or __bold__
  • link:[text](http://url.com/ "Title")
  • image?![alt text](/path/img.jpg "Title")
  • numbered list: 1. Foo 2. Bar
  • to add a line break simply add two spaces to where you would like the new line to be.
  • basic HTML tags are also supported

Tags:

×191
×36
×22
×3
×2

Asked: 11 Nov '13, 07:56

Seen: 10,217 times

Last updated: 09 Dec '13, 05:36

OR-Exchange! Your site for questions, answers, and announcements about operations research.