[OSM-dev] Tile server

Paul Norman penorman at mac.com
Sun Jun 23 00:02:38 UTC 2013

If you’re going to compare the SSD volumes to EBS, you need to compare it to provisioned iops volumes, which would be $400 a month for 4000 iops, and a minimum size of 400GB, for a total of $450/month. EC2 is more expensive for an osm2pgsql DB, which you’d want about 300GB currently and more in the future.


Just for comparison, Hetzner will rent you a 64GB RAM 6-core server with a 600GB SSD for 210 €/month, 100 € setup, or 32GB RAM, 4-core server with 2x 240GB SSD for 140 €/month, 150 € setup. OVH has a 64GB 4-core server with 2x 300GB SSD for 170€/month with no setup. Of course, all of these options should have far faster IO than a 4k iops volume.


If you’re doing dev work the OVH option is cheaper than EC2 if you’d need more than about a week of instance time in a month.


The conclusion you can draw from all of this is that if you’re doing prototyping or dev work, use an extract, have enough RAM and tune your postgresql for your use case. 


If you’re doing benchmarking this does not apply, but that’s a special case where you really need to give careful consideration to the workload and specifications of the machineyou’re using.


From: malcolm stanley [mailto:a.malcolm.stanley at gmail.com] 
Sent: Saturday, June 22, 2013 2:20 PM
To: Jeffrey Ollie
Cc: Paul Norman; OSM-Dev Openstreetmap
Subject: Re: [OSM-dev] Tile server


If you look here http://www.rackspace.com/cloud/block-storage/pricing/ you will see pricing for Rackspace SSD is $0.70 cents per GB per month.

Its hard to compare apples to apples as they include IO in the price.

Having said that, it looks like at that price a 500G volume for a world database would run about $350 per month.

a comparable Amazon volume would be far less @12.5 cents as opposed to $0.70, but then you have to pay IO AND you can only use them on larger instances, so you are going to cost up on that side.


Its almost like you need a case study comparing the offerings based on specific usage profiles. 


for instance, a dev instance on amazon is going to cost less (I **think**) because you would have low IO and low disk costs.

as you ramp the IO up and start to add performance to AMZ, that would probably change pretty quickly

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/dev/attachments/20130622/853115b5/attachment.html>

More information about the dev mailing list