[Beowulf] Has anyone actually seen/used a cell system?

Eric W. Biederman ebiederm at xmission.com
Tue Oct 3 00:54:07 PDT 2006

"Andrew Shewmaker" <agshew at gmail.com> writes:

> On 10/2/06, Vincent Diepeveen <diep at xs4all.nl> wrote:
> ...
>> Sir, You need to write a compiler blindfolded i'm afraid, unless you are
>> going to use software from others. But well you mentionned 2 guys can solve
>> all the problems...
> Vincent, you might care to know that you are telling this to Eric Biederman,
> the author of romcc.  While he didn't write it blindfolded, he wrote it under
> the constraint that it cannot use RAM -- only registers.  It isn't the compiler
> you are looking for, but rest assured that he understands this subject.

I at least have a clue.  That effort was in the ballpark of 3 man
months (give or take a man month), and was my first effort at an
optimizing compiler.

romcc routinely compiles 1000's of lines of C code (at a time) in a
production environment.

So if one developer can create a decent usable compiler in a man year
or so.  I find the assertion that it takes 1000s of developers to make
a very good compiler questionable.  My guess is that figure is a
at least an order of magnitude high.  Especially when you only
consider developers and not marketing, sales, testers, support
personnel and the like.

I think Intel's big advantage is that their developers can get several
years head start on supporting a new core or other new processor

Beyond that I find it hard to judge because it has been many years
since I had to fix a performance problem with something other than
an algorithm change, or coding that bit in assembly.


More information about the Beowulf mailing list