Electronics design moves to Linux (NewsForge) [LWN.net]

Electronics design moves to Linux (NewsForge)

Posted Jul 9, 2004 22:50 UTC (Fri) by dlang (subscriber, #313) [Link]

the problem is seldom that the programs can't be made to run on an 'unsupported' Linux, the problem is that when you do run into a valid bug in the program the first response from far to many vendors is 'we don't support that OS, switch to a supported OS' if the problem is intermittent this can be especially hard to track down.

Electronics design moves to Linux (NewsForge)

Posted Jul 10, 2004 1:30 UTC (Sat) by Duncan (guest, #6647) [Link]

Of course, the REAL problem is that the code isn't libreware in the FIRST
place, running on Linux or not. If it WERE libreware, one wouldn't be
dependent on the vendor for support, as one could go elsewhere or do their
own support.

If the hood is welded shut, it's STILL welded shut, whether the vehicle is
running on a proprietary railroad track or an open standards highway.
It's STILL crazy to buy such a vehicle, and users STILL suffer for it,
even if at times it IS the only currently available choice, thus, the best
choice available under the circumstances.

Likewise, the problem of all the different Linux distribs.. is only a
problem for those wishing to weld their hoods shut. If the hood is open
and the engine available for modification, it's not a problem, as those
that need it on other platforms ensure that it gets there, either by
paying someone to make the necessary changes, or by doing it themselves.

Duncan

Electronics design moves to Linux (NewsForge)

Posted Jul 10, 2004 7:04 UTC (Sat) by dlang (subscriber, #313) [Link]

you are mixing things up. there are programs that you can get the source for that the author still only supports on $favored_distro

I will admit that the problem is worse for commercial software, but try to go out and pay for support for a distro that's not in the top five and you will have a very hard time finding a company to support you.

for that matter isn't it all supposed to be about freedom? true freedom includes the choice to buy software if you want to.

Electronics design moves to Linux (NewsForge)

Posted Jul 10, 2004 13:14 UTC (Sat) by Duncan (guest, #6647) [Link]

> there are programs that you can get the source for
> that the author still only supports on $favored_distro

That's one of the blessings of open source. When you have the source (and
the license to use it as necessary, the part MS can't give), you don't
HAVE to depend on the author. You can do the support yourself, or
contract it out, on a consultancy basis if necessary. Pay the money
(which these companies are already doing.. it'd likely STILL be cheaper to
special-consultant an open source deal, given the tens or hundreds of
thousands some are paying for the software and support, and get a few
companies splitting the cost...), and someone WILL be willing to support
it.

> true freedom includes the choice to buy software if you want to.

Who said anything about not BUYING it? Remember, there's lots of
commercial software libre. Libre does NOT necessarily mean gratis, nor
would it be practical in this case anyway, as people spending $100k+/year
on a developer WILL want to ensure his software is supported, and
$10k/year/seat is NOT unreasonable, if he'd otherwise be idle or using
substantially less efficient design tools. I'm just saying it's a wise
idea to spend it on libreware, where they have the ability to go elsewhere
if necessary, building even MORE productivity security into the software
that developer is using, since it's no longer a single-point supplier.

Duncan

Electronics design moves to Linux (NewsForge)

Posted Jul 10, 2004 9:22 UTC (Sat) by phip (subscriber, #1715) [Link]

Libre EDA software is a difficult problem. The user/customer base for EDA software is pretty much limited to companies that already spend big bucks on design engineers and Fab facilities. These companies have no problem paying 10's of thousands of dollars for quality software and associated support (which is an entirely different world from the "support" you get in the consumer realm of commercial software). They are also reluctant to open-source any internally developed EDA software.

Much of the free EDA software that is available is developed by academic researchers who lack the interest and resources to develop a complete, polished product that would tempt companies away from the commercial EDA tools.

I think Icarus Verilog is taking the right approach - building a following of small FPGA developers who are below the radar of the big EDA companies. However, that approach will be much more difficult for the "back end" design tools (layout, DRC, paracitic extraction, etc.), because they are more closely tied to the fabrication process.

- Phil

Electronics design software as libreware?

Posted Jul 10, 2004 12:48 UTC (Sat) by Duncan (guest, #6647) [Link]

> Libre EDA software is a difficult problem.

That's why I allowed that it might be the best available /current/ choice,
but it's STILL not in theory the best choice, period. The problem of
closed source remains, even when it's the only realistic choice
(currently) available, that problem being the hood is welded shut, whether
or not it's the only vehicle available.

In such cases, it'd probably be wise for those companies spending tens of
thousands on quality software and support, to put a portion of it into
development of an open source solution, so they won't be stuck in that
predicament forever. That'd be the best choice for the future, even if
the current best choice requires purchasing that vehicle with the hood
welded shut.

Of course, that's the ideal.. Getting them to see that and do it.. But it
WILL happen. If they are spending THAT sort of money on it, it WILL
happen, even if it's a small niche market, because THAT sort of money is
just BEGGING for it to happen, eventually, and as the rest of the world
moves that way.. it WILL happen, it's just a matter of time. (Take for
example what the telecom companies are sinking into Linux, now.. A few
years ago, that would have been unthinkable as well, but it DID happen,
and IS happening. It's the SAME sort of money we are talking, I'd
imagine.)

Duncan

duncan, no

Posted Jul 10, 2004 17:20 UTC (Sat) by neoprene (guest, #8520) [Link]

The argument that EVERY piece of code has to be "libre" would preclude any commercial software for Linux. The larger Linux commumity will have no real need for a $100,000 EDA package. RPM's, DEB's work fine, not everything on this planet is compiled from source.

These narrow apps don't need a wide footprint of various platforms, one or two may be OK, the larger Linux community will have no need for these. The cost of an extra box for these expensive apps is very minor.

Some of the AWR's concerns about compatibility may be bogus, but they have a suspicion about their app not working right, and the money to test it out may not be there.

A seat of MicrowaveOffice costs about $24,000, I'll doubt they sell anymore if they port it to Linux, *unless* Linux offers a HUGE performance advantage***[see footnote]. The cost of an extra Win-harddrive to run this app may cost much less than $100, heck AWR may even toss it in for "free" :)

Your arguent that all apps has to be "libre" is not convincing enough.

***
If M$ is slow to scale up their OS to 64-bit with multiprocessors Linux may very well have an early start for some heavy lifting workstation apps.

Moreover, a 2-way of 4-way Opteron chock-full of memory will likely beat anything Sun has to offer.
http://store.sun.com/CMTemplate/CEServlet?process=SunStore&cmdViewProduct_CP&catid=108738
http://www.monarchcomputer.com/Merchant2/merchant.mv?Screen=PROD&Store_Code=M&Product_Code=80544&Category_Code=64Systems

Re: Electronics design moves to Linux

Posted Jul 12, 2004 6:47 UTC (Mon) by jvotaw (subscriber, #3678) [Link]

Re: support: thankfully, the bugs we've seen with EDA tools have generally fallen cleanly into one of two categories: either environment / OS issues, which we fix ourselves, or tools issues, which the vendors fix. So far we haven't had a case where a new revision of the tool takes three times as much memory as before, for example, or generates buggy output, and the vendor blames it on Debian. I've certainly seen that kind of problem with other software, but not thus far with EDA tools. Let's hope it stays that way.

Re: libre: I'd certainly like to see top-notch free EDA software, but it doesn't currently seem to exist and the amount of work it would take to replace the tools our engineers use would be prohibitive. It's more than a case of adding a couple of features to an already-existing package. Plus there's the issue of patent minefields...

Re: cost: I'm pretty insulated from software licensing costs, but I believe I've heard that some EDA tools cost upper six figures, per seat, for the initial purchase; figure 20% of that cost in successive years for support. Ouch. I think I'm in the wrong business...

-Joel

Re: Electronics design moves to Linux

Posted Jul 15, 2004 4:03 UTC (Thu) by JoeBuck (subscriber, #2330) [Link]

A state-of-the-art place-and-route tool that works for high-end ASICs will set you back $200K per seat. But then, the cost of turning the output of the P&R tool into a set of chip masks will set you back millions, and that $200K buys you application engineers who will fly in on planes to hold your hand if you have problems (particularly if you threaten to switch to the competition). And there are only a couple of thousand seats purchased worldwide, and this wouldn't change much if the P&R tools were free, since the cost of producing the masks would still be astronomical. Also, since masks are now used with UV wavelengths that are considerably longer than chip feature sizes, complex optical phase correction techniques are used to counter diffraction effects, and these techniques are all heavily patented (and these are really hardware patents, not software patents, because they cover techniques for producing a physical object, the mask, even if software does it).

That said, open source has made some dents in the EDA space, particularly in verification (e.g. the SystemC library). But for implementation, there are huge numbers of patents, and a ban on software patents wouldn't help open source much, because many of these patents deal with methods to produce a physical product, the chip. Even in a no-software-patents world, you might be able to use the code and simulate the result, but fabricating an actual chip would violate patents.

Re: Electronics design moves to Linux

Posted Jul 23, 2004 7:22 UTC (Fri) by xoddam (subscriber, #2322) [Link]

Certainly these patents are valid in the way business-method and
algorithm patents are not. Maybe the traditional patent period is
appropriate in this field, given the huge costs involved. But patents do
eventually expire, and when they do it would be very good to have
'logiciels libre' tools on hand. Perhaps the present tools vendors will
by then have decided to relax their licensing conditions anyway...