[racket] Engineering Tradeoffs of ANF transforms and the Stateless Server

That's a fine position in theory, but I doubt it works well in
practice even today.  When Jay talks about 3-4 GB, he doesn't mean
over the course of a conference -- that could sometimes add up in just
a few hours.  No amount of cheap RAM is going to combat that.

But, as I said, in an Ajax world there is much less need for the rapid
accumulation of continuations.

Shriram

On Sat, Dec 31, 2011 at 4:03 PM, Neil Van Dyke <neil at neilvandyke.org> wrote:
> Regarding paging... The only three situations in which I consider paging
> acceptable nowadays on workstations and servers, when a handful of dollars
> buys you a few GB RAM:
>
> 1. For resilience in exceptional/emergency situations, when the alternative
> would be an OOM kill or a crash.
>
> 2. When an application with massive memory requirements was designed/tuned
> specifically to work well with OS's virtual memory behavior.
>
> 3. Cloud and shared-hosting servers that don't care if performance blows.
>
> None of my personal machines even have swap partitions, and I don't recall
> using even half of my RAM.  In this screenshot of lower-right corner of my
> screen, it tells me I'm using 16% of 3GB RAM on this workstation that's
> running Linux, X, XMonad, PostgreSQL server, Apache, and several different
> Racket SCGI servers, in addition to my bloated email client:
> http://i.imgur.com/fCBBR.png
>
> --
> http://www.neilvandyke.org/
>
>
> ____________________
>  Racket Users list:
>  http://lists.racket-lang.org/users