On Mar 9, 2012, at 8:16 AM, Brian Mastenbrook wrote: > On 03/04/2012 12:20 AM, John Clements wrote: >> I have a bunch of students planning to run racket web-servers on a single >> machine. I just ran a few tests (on my mac, not on the target machine), and >> it looks like separate web-server instances chew up about 72M apiece, but >> that running 10 web-servers simultaneously on one process is only about 72M >> total. I have about 25 teams, so if they're all running a racket web server, >> that's going to wreck my poor 1G VPS. >> >> It seems like the right solution here is to run them all on one runtime. >> Moreover, with sandboxes, it seems that this should be eminently possible. >> I'd like to preserve their control over their own processes, though, so I >> thought I might provide a wrapper binary that passes a racket file over to a >> single central runtime to run in a sandbox, and that routes >> stdin/stout/stderr -- or something like it --- back to the user process. >> >> Has someone already done something like this? > > This would be a really cool feature of general utility. OS level containers > and virtualization systems are very popular these days; I think the same idea > applies at the language level too. > > Did you use multiple places in your test, or did you run everything within > the same place? It would be interesting to see how one racket runtime with > one place per user compared to multiple runtimes, both in memory and in > performance.
Just one place. I believe th… no, I agree. That would be a good test. John Clements
smime.p7s
Description: S/MIME cryptographic signature
____________________ Racket Users list: http://lists.racket-lang.org/users
