[OSM-talk-nl] OSMF fondsen & hardware

Stefan de Konink stefan at konink.de
Mon Feb 9 03:47:03 UTC 2009


Grant Slater wrote:
> Stefan de Konink wrote:
>>
>> No problem, I wonder who pointed you on this thread anyway. Please 
>> take this text as constructive, it is not intended to be written as 
>> troll, flame or whatever. It is written between 2:45 and 3:30am so... 
>> not the time these mails should be written anyway.
> 
> A few questions from the thread were asked on IRC  #osm. I followed up.

Oki :) some 'old' osm-nl people showed up here too :) So I was pleased 
that this was not a out of the blue statement.

>> Is this also the case for the Platinum sponsorship?
>>
> Slow in paying, yes likely. Don't most companies wait till 89.5 days. ;-)

Actually the Dutch max is one month. After that 'state-interest' is 
added to any bill made, plus any extra admin costs. So over here usually 
everyone pays in the 3rd week. Never the less, we are talking here about 
a well known OSM friendly Platinum sponsor; imho paying in advance 
should never be a problem.

>> Are the PIII servers still connected? If so, we might have a job for 
>> them :)
>>
> 
> No, the 2x PIII servers are not currently connected. Let me know what 
> you need them for and I can try get them connected.

As you might know we have implemented a 'speedy' tile server, based on a 
very light, high performance open source, webserver. I would like to geo 
balance requests from the UK to the UK and NL to NL. Especially if we 
are going to render the world soon.

Some simple GIS stuff with GeoJSON/KML is one of the new things we would 
like to introduce with public transport. Hover over a busstop, and you 
will see all the routes connecting to it.

> (Assistance always appreciated). I think 2009 is going to be a 
> large growth year for the project.

I think the best thing would be to keep you in the loop, and try to set 
up some geo balancing in the non API resources. Helping out is never an 
issue :)

> The current DB server is an up-spec'ed desktop machine, it's amazing it 
> has lasted us this long.

I totally agree. When I wrote my last benchmark paper I was very 
surprised I had the actual hardware under my desk that the 
Ruby-API/Database was running on. But never the less, we are also 
looking at a software bottleneck. Just blaming the proper working 
hardware is not my thing. For the record, I did bench OSM also on 4x as 
big hardware (Dell 1950 2x Quad Xeon 32GB) also at that thing 
memory/disk was the major bottleneck on random requests.

>> Now I sincerely hope OSM gets a freaking Sun X4540, with 64GB* of RAM 
>> and decent 48 disks array. And if someone is really smart, maybe ask 
>> if OSM can get it for free, instead of buying it with this new money. 
>> You can afford it anyway now :)
> 
> This is my intention. Got any good contacts at Sun or HP? ;-)

I have a X4540 in my development rack to benchmark.
http://xen.bot.nu/benchmarks/nfs.html
(OpenStorage was done without Jumboframes though, that might indicated 
why it output is that bad)

Today I heard even at the OSM-NL folks there is proper contact with Sun, 
if you do go for this X4540, do it proper. You could use it as an 
OpenStorage box (appliance). But probably the best value for no money is 
installing OpenSolaris. I think you could even get additional 
performance if you compiled MySQL yourself under a Gentoo/Solaris x86_64 
environment. But this box basically eats everything as biscuits.

Since more disks are available SAS is not really a requirement.

>> * at this amount of memory, ditching MySQL for something else might be 
>> an option.
> 
> A lot of work has gone into making the 0.6 API backend compatible with 
> PostgreSQL, unfortunately no promises yet. Maybe 0.601 ? :-)

If the next update is after june 30 there will be something better ;) 
Otherwise I'll not get my M.Sc. ;)


Stefan




More information about the Talk-nl mailing list