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

Re: continuations in the real world?

Ok, then how do you deal with failover?
On large applications session afinity means that a
user is much more vulnerable to the failure of
their specific machine.  Obviously this type of
robustness matters more for some applications than
others, but if you are doing stock trading, and
you have an MTBF of e.g. 2 years and you have 100
machines, in practice you have failure pretty


On 26 Mar 2002, Paul Q.Graham wrote:

> It depends on the granularity at which you divide stuff
> over multiple machines.  At Viaweb each user was sent
> to a given machine after logging in, and for the rest of
> their session talked to that machine.  When you have numbers
> of users/machine in the hundreds or thousands (and you
> have a weird application if the numbers are lower) the
> sample size is large enough to dwarf variations in loads
> between users.  --pg
> --Mario Latendresse wrote:
> >
> > Avi Bryant <avi@beta4.com> writes:
> >
> > > session state.  For those who use it, even (especially?) those who know
> > > nothing about continuations, it's not just of academic interest to be able
> > > to layer normal control flow over the HTTP request/response loop and get
> > > transparent backtracking to boot.  I know from personal experience that
> > > the use of continuations has a profound simplifying effect on the
> > > architecture of large web applications, and from what I'm hearing from
> > > early users of the framework, they're discovering the same thing -
> > > smalltalkers may not be used to dealing with full continuations, but they
> > > know good magic when they see it.
> >
> > If you scale an application over several machines, the use of
> > continuations require a distributed implementation of the underlying
> > runtime system. Without such a distributed implementation, which is
> > the current state for all functional implementations I am aware of,
> > this lack of scalability is a major problem. May be you have a
> > different solution or indeed such a distributed runtime system?
> >
> > Mario

S. Alexander Jacobson                   i2x Media
1-212-787-1914 voice                    1-603-288-1280 fax