[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: APT-like system for BOINC



>>>>> "Aaron" == Aaron Lehmann <aaronl@vitelus.com> writes:
    Aaron> You mentioned that all binaries would be signed. Would
    Aaron> you exempt unofficial URL's from the cryptographic
    Aaron> validation?

Sure, that would be reasonable.  I don't know what the policy for
third-party application servers would be...

    >> (4) System administrators?
    Aaron> 
    Aaron> I don't how this should really affect system
    Aaron> administration. I've run several distributed computing
    Aaron> projects, all under a user-level account. The only
    Aaron> system-level support that seems merited is an init
    Aaron> script and a dedicated user account, and both are
    Aaron> optional. Packages (deb/rpm/etc) would automate this
    Aaron> for people who don't want to bother.

Okay, I'm not sure either if anything special needs to be taken
into account.  Various people claiming to manage hundreds to
thousands of computers running SETI@home have complained about the
BOINC system.  One argument has been that IT at large
organizations need to do a lot of internal stability testing
before installing any software on user desktops, and automatic
updates will subvert that.

    Aaron> Perhaps whenever a client checks into the server to get
    Aaron> work or report results it could be notified of the
    Aaron> latest version of the binary in question. How to react
    Aaron> to a newer version could be a user setting.

Version requirements are a bit tricky.  Currently each workunit
has an application version associated with it so you need a
specific version to process this.  (This works fine right now.)
However, if we let people compile their own applications we will
have to support some kind of version range so that people have
some time to compile/update.


-- 
Karl 2003-12-31 19:37



Reply to: