Vat Location Service
Mark S. Miller
markm@erights.org
Mon, 11 Jan 1999 14:25:18 -0800
At 11:19 AM 1/7/99 , Bill Frantz wrote:
>MarkM asked me over dinner last night where we could bring up a Vat
>Location Service (VLS). The basic requirements are a 24/7 Internet
>connected machine capable of running a Java virtual machine.
>
>Well, on the way home, my wife Peri mentioned the magic word, HTTP. It
>occurs to me that if we implemented a HTTP/CGI based VLS, then we could put
>up a VLS on any ISP that supported user CGIs. Doing this would require
>some changes in DataComm to recognize a URL for the search path and run the
>VLS under HTTP protocol, but I think that is quite doable.
>
>Reactions?
[?] Does this fall out?
Once your http-tunnelling code is completed, could we use it to trivially
layer the VLS protocol on top of HTTP? Flipping it around, might your HTTP
tunnelling work enable people to inefficiently put persistent Vat-based
services up at their ISPs as CGIs?
With what permissions do ISPs normally run CGIs? Would they enable a
Vat/CGI to checkpoint itself between requests? Note that the checkpoint
file contains secrets that must remain inaccessible to other unprivileged
users of the ISP.
Can a process spawned by a CGI stay around after the CGI request is done?
Can it establish its own TCP/IP communications, and can these stick around?
For the moment, we have VLS code that should work, and descends from VLS
code that worked very well under difficult real world conditions (EC
Habitats (aka Microcosm) beta). However, we aren't running any VLSs
anywhere, so the standard E install contains no default search path. Until
this is fixed, we cannot begin to do E-based distributed computing. (It's
like an internet without anything filling the role of DNS, and without
stable IP addresses!) We need a small number of volunteer machines. The
VLS is a java process that must be restarted automatically, and must be
able to listen on a stable TCP/IP port accessible from the open internet.
Assuming we can get these volunteers, I'd rather go with the protocol we've
got for now, rather that worry about HTTP.
Eric Messick has generously volunteered to host a persistent VLS on one of
his machines. Thanks! We just need a few more. Sites outside the US, but
with low latency connection into the US, would be especially nice. High
bandwidth is *not* needed.