🚀 go-pugleaf

haertel@ichips.intel.com (Mike Haertel) writes:
] Dennis O'Connor~ <doconnor@sedona.intel.com> wrote:
] >Admit it : being order-neutral is
] >slower.  Ask your typical customer wether they'd rather have a
] >byte-order-neutral OS that runs slower, or a machine-order-dependant
] >OS that runs faster, and which do you think they will buy ?
]
] More seriously, I completely disagree with Dennis.  I don't see
] any great advantage to standardizing a byte order in an OS, since
] no matter which order you standardize you disagree with about half
] the world.  (Internet protocols are big endian, most on-disk data
] structures in PC applications are little-endian.)
]
] Since this is comp.arch, does anybody have any actual performance
] numbers for the costs of byte swapping?

It's one thing to have some small amount of code in your OS that knows
it must byte-swap everytime (like a TCP/IP stack on an Intel Architecture
computer).  It's quite another to degrade the performance of your
entire OS by making the entire OS byte-order insensitive.

Besides, at a mere 10 or 100Mbps, the network is already pretty slow
(compared to current CPUs that is). So what if it's a little slower ? 
If you _really_ care (and some do) get an intelligent network interface
card to do the bottom of the TCP/IP stack for you.

Ain't that just like an Intel engineer to propose throwing more
processor (or in this case, more processorS ) at a problem ? :-)
--
Dennis O'Connor                            doconnor@sedona.intel.com
i960(R) Architecture and Core Design       Not an Intel spokesman.
TIP#518					   Fear is the enemy.