To avoid the customer having to get a fixed IP address (or having to build a dynamic IP update client into the gateway), configure their router to give a fixed IP address to the gateway, and to forward a port to the gateway: we could have all data exchanged by periodic requests from the gateway. “Got anything for me?” or “Here’s what I’ve got” every X seconds, minutes, etc.

The question is: how often can we poll the server? Is this a sustainable, scalable architecture? What if there are 100k aiospheres in the world? What if they each polled the server every ten seconds? Is this a lot of data for the internet? For a web host?

• Say the communication in each direction is 512 bytes or 1k total.
• 1024 bytes x 6 times a min x 60 min x 24hrs x 30 days = 265MB per unit, per month.
• If we sell 100k units, our servers would see a monthly load of 26TB per month.
• Cheap shared server web hosts (HostGator) allow 6TB per month for $7, and unlimited for $12/mo.
• They also sell dedicated servers with 1.5TB per month for $175.
• JaguarPC has dedicated servers with 1.2TB for $120.
• Rackspace offers 200-500GB bw.

This weekend I read Embedded Ethernet and Internet by Jan Axelson. My understanding of Ethernet is still pretty hazy, but I did learn that a TCP packet is 20-50 bytes (typ closer to 20). So a “got data?” query, and a “no” ack, could be a total of about 50 bytes. This reduces my initial rough bw calcs by more than order of magnitude. Does this bring the scheme more into line with appropriate network use?