Re: REST impact paper

It was thus said that the Great J. Andrew Rogers once stated:
> 
> > On Sep 8, 2017, at 10:44 PM, Stephen D. Williams <[email protected]> wrote:
> > 
> > Proprietary code that you can't get to, probably because you can't afford it, doesn't matter.  With the cost of commodity hardware, per-core style of pricing may not be competitive with only a 2-5x improvement. 
> 
> 
> Unrestricted source licenses to things I would want to license (because
> they are awesome) are typically an inexpensive one-time cost, usually less
> than the cost of an engineer to write it poorly. 

  Wow, are you lucky.  At work, we paid a high six figures just to *license*
(no source code initially [1]) a network stack and it's a piece of shit. 
It's hard to configure (vwl mprd cmmnd ln ntrfc strght frm th 60s) and if
you look at it funny, it pukes [2].

  I was amazed when I saw the code.  Not at how elegent it was but more of
how horrible it was.  Call an external function and oh look, a linear scan
for data [3].  Oh, and an internal call does tha same linear scan.

  Yeah, I'm haven't been impressed with third party proprietary code.

> I’ve never seen it
> structured on a per-core/per-seat basis. 

  Doesn't Oracle charge per CPU?

  I've certainly heard of programs that charged by the core/seat.

  -spc (Then there was my one experience using SCO Unix ... )

[1]	We eventually got the source code, but only because it's now it's an
	"end-of-life" product from the firm who has long since outsourced
	support for it to a firm in India.

[2]	Mainly because if the network peers don't like *anything* about the
	connection, including too much traffic, they can drop the
	connection.  Too many errors?  Connection dropped.  Too much? 
	Dropped.  Too little?  Dropped.  Can't tell the difference between
	African and European swallows?  Dropped.

[3]	The API originally took an integer, a Unix file descriptor, way back
	in 1987 when the code was first written.  At some later point, it
	became clear that they needed to track more information about the
	descriptor and instead of biting the bullet and redesigning the API,
	kept it, and now do a linear scan to find the associated data for a
	descriptor.

	High performance my ...

_______________________________________________
FoRK mailing list
http://xent.com/mailman/listinfo/fork