Sharedband Hits Seattle

said by tiger72:

Why not just get these connections separately and use a dual-wan router on the client side? What's the point of a middle-man?

said by dvd536:

Will this take 4 1.5mbps DSL lines and turn it into one 6mbps pipe or is it simply load balancing? if the latter then *yawn* as dual wan routers have done this for years

said by dvd536:

Will this take 4 1.5mbps DSL lines and turn it into one 6mbps pipe or is it simply load balancing? if the latter then *yawn* as dual wan routers have done this for years

said by espaeth:

MLPPP is just an IP tunnel, so no ISP support is required. (except for your ISP allowing you to have multiple broadband hookups at your location)

You create an IP tunnel (like a VPN without the encryption) on each of your Internet connections to their infrastructure, and you do per-packet load balancing to "spray" the traffic across all of tunnels to get your aggregate bandwidth.

said by wifi4milez:

Yes, however without ISP support you are doing nothing more than load balancing.

DSL#1  10.10.10.1  <----->    MLPPP agg (Shareband) 20.20.20.1
DSL#2  10.10.20.2  <----->    MLPPP agg (Shareband) 20.20.20.2
 
MLPPP Bundle 20.20.1.1 <--->    MLPPP Bundle 20.20.1.2
Where the MLPPP bundle is built from PPP tunnels across DSL#1 and DSL#2
MLPPP bundle uses Shareband IP space inside the tunnel, not your ISP's.
 

said by espaeth:

Let me try this again. It's like this:

DSL#1  10.10.10.1  <----->    MLPPP agg (Shareband) 20.20.20.1
DSL#2  10.10.20.2  <----->    MLPPP agg (Shareband) 20.20.20.2
 
MLPPP Bundle 20.20.1.1 <--->    MLPPP Bundle 20.20.1.2
Where the MLPPP bundle is built from PPP tunnels across DSL#1 and DSL#2
MLPPP bundle uses Shareband IP space inside the tunnel, not your ISP's.
 

Today when you get 2 broadband connections you get a unique IP per connection, which is why you are limited to the bandwidth of one connection at a time. (each TCP session can only have a single source IP)

Shareband builds a Point-to-Point Protocol (PPP) tunnel on each connection and then bundles the tunnels together in a MultiLink PPP (MLPPP) tunnel so that the MLPPP virtual interface on each side has a single IP address. Each packet destined for the bundle is usually split between each tunnel in a round-robin fashion so that are get the effective throughput of all of the connections minus the PPP overhead.

You can build the same thing yourself by getting a couple broadband connections at your house and building a MLPPP session to a device (be it router or server) that you colo at a nearby facility. The reason that it needs to be close is because all packets need to run to the aggregator and then out to the public Internet. If you put your MLPPP termination too far away, your performance will actually decrease due to the extra round-trip latency you pick up.

It should also be noted that this generally won't work with unlike pairs. Ie, you can aggregate multiple cable or multiple DSL connections, but not a cable and a DSL connection. The network performance difference between the two networks will result in severely out-of-sequence packets arriving at the MLPPP aggregator and performance will be crap.

said by Guspaz:

TekSavvy, a Canadian national ISP serving roughly 85% of the country's population (the four largest provinces), offers unofficial support for MLPPP on their PPPoE-based DSL connections.

Numerous people with the technical skills to set up a *nix box are enjoying connections their bonded DSL connections (currently bonding two 5mbit down 800kbit up DSL lines).

Their routers support either packet round-robin, or splitting the packets. Either way, the effect is the same; any given IP stream only sees one extra-phat tube, as TekSavvy's router re-combines things on their end.

said by Guspaz:

As has been said earlier, there is one way to do it without ISP support; create two PPP tunnels to a remote box (operated by sharedband), and use mlppp. The downside is that you're adding extra latency to your connection, since your internet connection now effectively terminates at sharedband's box. Your public-facing IP would also now be a sharedband IP.

said by Guspaz:

It shouldn't be a problem. Any ISP that is a wholesaler of Bell Canada's DSL services (TekSavvy included) has about $9 to play with as far as bandwidth, equipment, staff, etc.

Bell charges $20.50 for the line, and $1300 for the GigE to connect the DSL clients to the ISP's network (this is not transit, just a connection to Bell's ATM network, IIRC). That leaves about $9 per customer to pay for everything else when you consider that TekSavvy charges $29.95 for their normal 5/800 service.

In that $9, TekSavvy is able to provide 200GB/mth of bandwidth (5mbit DSL), and cover all other expenses, and still turn a profit. I don't see why a company like Sharebond can't make a profit on $10 or $20 per customer, as their actual bandwidth costs would be lower due to average usage. TekSavvy has proven that you can act as an ISP on those kinds of margins, turn a profit, and still manage to become the highest rated DSL ISP in North America (by DSLR's rating) :P

said by Guspaz:

That's the thing though, Bell wholesalers are responsible for maintaining their own connection. The $20.50 connection fee and $1300 GigE to Bell only gets the user's packets to the ISP (TekSavvy)'s network. From there, the ISP is responsible for getting the packets out to the internet at large.

They currently have six GigE lines to various providers (Cogent, Teleglobe, Internap, Peer1, etc) for that purpose.

My point was that if you eliminate the two Bell-related charges, and possibly some of the equipment costs required to connect to Bell, you're left with TekSavvy's $9 per month or so. Inside of that, they're able to cover all expenses and provide 200GB of actual IP transit. For $19 ($39.95 to customer, $19 is roughly left over after Bell takes their cut), they're able to provide unlimited services, targeted at users who use over 300GB per month (300-500 is probably typical).

If TekSavvy can make a profit, I fail to see why Shareband, with very similar costs once Bell is removed from the equation, couldn't do the same thing.