[evla-sw-discuss] [Fwd: Re: Accelerated Technologies "Nucleus" ROTS]

[Bill Sahr]

-------- Original Message --------
Subject: Re: Accelerated Technologies "Nucleus" ROTS
Date: Wed, 12 Dec 2001 11:58:51 -0700
From: Bill Sahr <bsahr at cv3.cv.nrao.edu>
Organization: National Radio Astronomy Observatory
To: Kevin Ryan <kryan at zia.aoc.NRAO.EDU>
CC: bsahr at zia.aoc.NRAO.EDU, ghunt at zia.aoc.NRAO.EDU,
gvanmoor at zia.aoc.NRAO.EDU,kryan at zia.aoc.NRAO.EDU,
browen at zia.aoc.NRAO.EDU,Wayne Koski <wkoski at cv3.cv.nrao.edu>,George Peck
<gpeck at cv3.cv.nrao.edu>,Ken Sowinski <ksowinski at cv3.cv.nrao.edu>
References: <200112121452.HAA20164 at ozone.aoc.NRAO.EDU>

re Kevins comments on the cost of the ATI Nucleus software
for the EC-1 candidate MIB "sytem on a chip":

When I was doing the cost data sheets for the EVLA project,
I priced out a VxWorks license for the PowerPC architecture.
Including the per project & per seat costs, 5 seats (4 seats
cost more than 5 becuse of volume discounts) the cost was
$48.5K.  Using Bruce's figures given below, 5 seats for the
shell, plus the nucleus, plus the network stack runs $34K.
The $34K figure does not include money we might save from
items supplied by the EC-1 vendor, and it also does not 
include money we might have to spend for additional, possibly
per seat, items such as a more sophisticated debugging
capability.

So, where does that leave us ?  The EC-1 uses an Intel core.
Our present VxWorks license is for 68XXX.  To use VxWorks
in the EC-1 would require purchase of an architecture license
for the Intel family.  No gain there.  As a matter of fact,
it's a loss since the VxWorks license for the Intel architecture
would probably be about the same $48K we would have to pay
for the PowerPC license as compared to $35K-$40K for the ATI
Nucleus EC-1 software.  Additionally, we would probably
still have to port VxWorks to the EC-1 since I doubt they
have a "ready to go" port for it.  A double loss.  More money
& more of _our_ time and manpower for VxWorks.  Finally,
we would still not have a PowerPC license, so we would either
have to purchase an additonal VxWorks license for the PowerPC
architecture or forgo the use of PowerPC CPUs for our real time
boxes, a step I am loathe to take since the specs for the
PowerPC are an order of magnitude better than the specs for
Intel architectures w.r.t. to characteristics important for 
real time work.

OK.  So why not use a PowerPC core for the MIB ?  First, I 
do not think that fundamental design choices should be constrained
by a $40K cost on a $75M project.  That sort of logic strikes me 
as false economy.  We can find the money.  Second, if we chose a 
PowerPC core for the MIB and purchased a VxWorks license for 
PowerPC we would still be faced with the task of porting VxWorks 
to the chosen MIB core/processor.  I do not want to see us involved
with that sort of effort, especially w.r.t. to a "system on a
chip" configuration rather than a more straightforward CPU
board, unless it is _absolutely_ unavoidable.  Such efforts are
notorious time sinks.  Even a few months of the time of one
staff member would be too much time.  It would be preferable
to spend the $40K for the ATI Nucleus software.  Also, such
efforts often have not entirely satisfactory results - some
possibly show stopping intermittent bug or bugs crop up and
take months more to run down.  I see it as a higher risk path
not justified by the money saved, which would threaten important
components of the project that are already on tight timelines.  
If the issue were $300K rather than $30K, I might feel differently, 
but it's not $300K. 

The only other possible win scenario, w.r.t. the economics, that I 
can foresee at this time is to proceed with the purchase of the
ATI Nucleus software for the MIB (if an Intel core is chosen) and 
consider the use of an open source real-time linux solution for the 
computing system real-time boxes.  Such a step would require _very_ 
careful consideration since the same logic of false economy might 
well apply.

In summary, I don't think the cost of the ATI Nucleus software
is sufficiently high to justify extraodinary efforts, higher
risk of schedule slippage, or the allocation of additional
man-months on our part.  Yes, it rankles to pay 10s of thousands,
but we do have a budget, and, at least at this time, we are much
more hard pressed for man-months than we are for dollars.

Dissenting opinions are welcome.  Agreement will also be well
received.

Bill


Kevin Ryan wrote:
> 
> I think, that at those prices, we should give WindRiver serious
> thought.  Saving $tens of thousands might justify the extra work
> and time of learning and setting up a new OS, but if the prices
> are nearly the same, we might be better off with what we know.
> 
> Unless, of course, this new OS offers other advantages over VxWorks
> that would justify its use.
> 
> Kevin
> 
> >
> >
> > The kernel of the OS runs $12495 (list)
> > the network stack runs $14995
> > per seat items i.e. shell $1300
> >
> 
> >
> > They claim to be a direct competitor to WindRiver and their OS features
> > are very similar. They
> > claim to have highly rated tech support and are #2 in the RTOS
> > marketplace.
> >