[OSM-dev] How to earn undying fame

[Nic Roets]

On Mon, Jun 21, 2010 at 5:17 PM, Tom Hughes <tom at compton.nu> wrote:
> On 21/06/10 16:00, Nic Roets wrote:
>
>> The result of the uservoice survey[1] is in: The community wants
>> routing on our website more than anything else.
>
> Like we didn't know that already.

Tom, you're in London and is an OSM insider. Even Steve Coast
indicated that he finds uservoice a useful tool.

I feel very out of the loop here in South Africa. I don't know that
the supply and demand for routing server software is and that's the
main reason why I have spent very little time on the routing back-end
during the last 20 months.

>> And most of the work has already been done. An opensource routing
>> engine that can do the job already exists [2]. The routing database
>> can be updated weekly. The hardware requirements are quite reasonable
>> (64-bit processor with tens of GB of RAM). If a new server is needed,
>> I'm sure we'll get the money in fairly quickly.
>
> I'm not sure "tens of Gb of RAM" is that reasonable, though it does rather
> depend how many tens you mean. It's also rather important to know how fast
> that memory requirement is likely to grow.
>
> Equally I would like to have some idea of how many routes/second you believe
> this engine to be capable of on that hardware.

It depends on how long each route is and how dense the road network is
in that area. So I must admit that I can't tell you because I don't
have access to a such a machine.

The amount of time and RAM used is proportional to the square of the
distance between the 'from' and the 'two' marker. So the load can
easily be managed by placing an upper limit on it.  So when we see an
increase in the load (e.g. TV show), a small reduction in that upper
limit will solve the problem.

> No. Whatever solution is adopted should absolutely not involve making a copy
> of the web site. It should be an addon to the web site, not a copy of it.

Like Kai said: A demonstration copy that will be merged with the main site.

> Updating the database weekly might not be able to meet people expectation like the instant rendering
> that we have.

AFAIK Google Earth takes a month to update their community layer. So
weekly updates will be a good start.

> In case someone would try and do it, it would probably be a good idea to make the frontend code as
> backend agnostic as possible.

> Well if somebody can provide a backend that returns a GPX or KML or something then I'm sure
> somebody will hack up the rails port to be able to display those routes.

You can look at the current "cgi" spec. It's richer than plain GPX,
but still quite simple. Using sed or awk to change it to GPX is quite
straight forward.

(Thanks Lambertus for answering the other questions).