From: billwolf%hazel.cs.clemson.edu@hubcap.clemson.edu (William Thomas Wolfe, 2847 )
Subject: Re: Compilation bottlenecks
Date: 25 Mar 90 00:45:31 GMT [thread overview]
Message-ID: <8489@hubcap.clemson.edu> (raw)
In-Reply-To: 6000006@hpcupt1.HP.COM
From jamiller@hpcupt1.HP.COM (Jim Miller):
> I know of customers which *will not* recompile, due to the
> impact on the time their operation -- one just better find a better way.
> They are not using C.
> jim - i'm not even sure these are my opinions - miller
Although this statement _seems_ to imply that the reason has
something to do with the computational complexity of compiling
a language in which there is considerably more compiler analysis
being done (such as Ada, in which very large projects have often
taken precisely this approach), in fact the REAL (tm) reason is that...
MOST ADA COMPILATIONS ARE LIMITED BY I/O CAPACITY, NOT BY CPU SPEED!!!
I heard at a conference recently of one project for which the time
required to recompile a very large software system went from around
ten days to around two or three days, simply by installing enough
I/O channel capacity to overcome the I/O bottleneck!!
Ada is doing a lot of work with the program library's database in order
to keep track of the relationships among the various compilation units,
which is very useful in terms of tracking dependencies and is exactly
as it should be. But it gives Ada compilation the basic characteristics
of any database-oriented application, and having the appropriate hardware
(such as the tremendous I/O capacity of typical IBM platforms) seems to be
the single most important key to overcoming this situation, and giving back
to the very large projects their ability to recompile over the weekend!
Bill Wolfe, wtwolfe@hubcap.clemson.edu
parent reply other threads:[~1990-03-25 0:45 UTC|newest]
Thread overview: expand[flat|nested] mbox.gz Atom feed
[parent not found: <6000006@hpcupt1.HP.COM>]
replies disabled
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox