Re: Floating point usage in a MIB module

[Andy Bierman <abierman@cisco.com>]

At 05:45 PM 1/30/2004, David T. Perkins wrote:
>HI,
>
>I agree with the distictions that Andy as pointed out below.
>I just don't find the cost factors to be excessly high
>in a system without floating point hardware. We continue
>to respectfully disagree on this issue. Probably a supplier
>of float emulation software can provide detailed numbers of
>the code footprint size. And, of course extensize
>use of float for management instrumentation in a system
>without hardware float support is going to bog down
>the CPU. But you all know these tradeoffs....

yep, and that's why it's highly unlikely float emulation
software would be added to our routers, just for SNMP
(and maybe just a couple of MIB objects at that!).

However, I don't really mind if a float data type (or TC)
is introduced, if it's done in a manner similar to
Counter64 (e.g., Counter32 pair).  For each MIB object
of type float, there should also be one or more 'legacy'
objects that provide an INTEGER representation of the
number (or best approximation).

Andy



>On Fri, 30 Jan 2004, Andy Bierman wrote:
>> At 05:09 PM 1/30/2004, David T. Perkins wrote:
>> >HI,
>> >>....
>> >I truely believe that floats are important. And even though
>> >the last time we talked about this, and Jeff Case and Andy Bierman
>> >were againest it due to the costs, I understand their concern
>> >but do not agree with the extent of the costs, and believe the
>> >benefits are quite great.
>> 
>> I think we should distinguish between floating point
>> operations and floating point representation.  It's
>> not difficult to add code to an agent to encode
>> an (internal) integer as a float.  It is difficult
>> to add float arithmetic support to an agent that
>> doesn't have any such code already.  
>> 
>> >>...
>> >Regards,
>> >/david t. perkins
>> 
>> Andy
>> 
>/david t. perkins