[erlang-questions] Lets build a stock exchange!
Thu Nov 15 15:57:13 CET 2007
Oh, okay. I am sorry to seem a bit slow.
- You want something interesting enough to induce a bunch of us to
run EC2 instances in order to test a humongous Erlang+EC2 cluster.
- It has to be interesting to you so that you get it done.
- It does *not* need to be production quality.
> On Nov 15, 2007, at 12:55 PM, G Bulmer wrote:
>>>> (I realise your focus
>>>> is a book, but wouldn't you like millions of people to use the
> No, I wouldn't. The software I will create for the book is just
> educational software. I will use this software to demonstrate
> Erlang features and to try to build a huge EC2 cluster.
> I will likely develop the software for millions of people to use
> after I write the book. .... There's not enough time or pages in
> the book to accomplish such a feat.
Okay, I understand, and can agree that designing and building a
usable stock exchange, at the same time as writing a book, may be too
much work in the time window available.
>>> Further, I am not sure how low the latency will be; I can't ping
>>> amazon, but when I ping google I get round-trip averages over 50
> I think latency here is the delay introduced by the software
> itself, not the delay introduced by the internet.
Well, Amdahl's law says the largest component of performance
dominates optimisations ...
So, is your plan to run the whole thing within EC2?
>> My problem is I don't see why *I* would run a stock exchange
>> application, so I don't see why *I* would run Amazon EC2 instances.
> You don't need to run a stock exchange application. You may want to
> add your instance to the cluster to take part in the Erlang
> scalability test.
Ah, ha! I think I've got it!
> .... May I suggest that you take up a parallel effort and describe
> how to build auctions on EC2 in a series of blog posts? The
> software you develop will be commercially viable and useful to you.
> It will serve as additional proof of viability of Erlang on EC2.
I am not passionate about auctions, it just seemed more likely that
people would use it, and hence it'd get a large number of instances
being 'thrashed' by live traffic. I now understand that is a 'non-
goal'. You only need something big for a 'cluster-fest'.
Maybe I will wait for the extremely-large-scale Erlang+EC2 cluster
issues to be solved.
That I am interested in!
More information about the erlang-questions