Great question about the routing. Will they create a new routing protocol. They will need unique routing protocols. I did a short search, but didn’t scour Reddit. I see early research and a pilot for IRIS from Cisco. Early versions don’t appear to be scalable to meet the Starlink architecture. This will be one of and potentially the largest commercial network in the world. Once the “starlink laser cloud” is running it would have a significant advantage for high frequency traders hedging between markets, and all cloud service providers. Bezos os in a losing position, but will probably spends 10’s of billions to try to be a close competitor to support AWS. SpaceX appears to have a 3 to 5+ year advantage on any competition. Figuring out the logic to make this work well will be up there with the pioneers of the internet.
I'm guessing here...Unperturbed orbits are very reliable, and every satellite could have the database for the current Starlink fleet, and a routine for keeping up to the microsecond positioning of all near neighbours. Reasonable pointing from that is just maths. Maybe there is a homing routine to center the beam once acquired.
Quote from: DistantTemple on 09/07/2020 05:42 pmWill SX develop a new in-house routing layer? SRP, Starlink Routing Protocol? Existing networks do not have their nodes constantly moving, reshaping the routes via changing routers. Also Since SX (we assume) plans to sell services to stock exchanges, international finance, the military, and "secret service", some way of hiding the routing from any kind of snooping....I know too little about even TCP/IP to go on..... Just everyone else has to fit into the existing internet..... backbone, but SpaceX is effectively making a new network. At first it will rely heavily on the existing... but once the ISL are up, Many connections will be entirely within SL. And even those that only start or end within SL (subscribers/clients) may spend the majority of their journeys within SL, and only need (translation) to the local internet for the last few 10's of miles.TCP is out dated and used by HTTP/1.1 and HTTP/2 (slated for depreciation). HTTP/3 uses QUIC (Quick UDP Internet Connections) and is the successor to TCP.Source: https://quicwg.org/base-drafts/draft-ietf-quic-http.html
Will SX develop a new in-house routing layer? SRP, Starlink Routing Protocol? Existing networks do not have their nodes constantly moving, reshaping the routes via changing routers. Also Since SX (we assume) plans to sell services to stock exchanges, international finance, the military, and "secret service", some way of hiding the routing from any kind of snooping....I know too little about even TCP/IP to go on..... Just everyone else has to fit into the existing internet..... backbone, but SpaceX is effectively making a new network. At first it will rely heavily on the existing... but once the ISL are up, Many connections will be entirely within SL. And even those that only start or end within SL (subscribers/clients) may spend the majority of their journeys within SL, and only need (translation) to the local internet for the last few 10's of miles.
Quote from: russianhalo117 on 09/07/2020 05:51 pmQuote from: DistantTemple on 09/07/2020 05:42 pmWill SX develop a new in-house routing layer? SRP, Starlink Routing Protocol? Existing networks do not have their nodes constantly moving, reshaping the routes via changing routers. Also Since SX (we assume) plans to sell services to stock exchanges, international finance, the military, and "secret service", some way of hiding the routing from any kind of snooping....I know too little about even TCP/IP to go on..... Just everyone else has to fit into the existing internet..... backbone, but SpaceX is effectively making a new network. At first it will rely heavily on the existing... but once the ISL are up, Many connections will be entirely within SL. And even those that only start or end within SL (subscribers/clients) may spend the majority of their journeys within SL, and only need (translation) to the local internet for the last few 10's of miles.TCP is out dated and used by HTTP/1.1 and HTTP/2 (slated for depreciation). HTTP/3 uses QUIC (Quick UDP Internet Connections) and is the successor to TCP.Source: https://quicwg.org/base-drafts/draft-ietf-quic-http.htmlTCP and QUIC are both Transport layer protocols, one layer above IP, which is the principal Network layer protocol of the TCP/IP stack. The Network layer sends packets around the network, routing them to their destinations. So IP is the layer that does routing. TCP is a transport layer protocol, one level up. It adds error checking, retransmission, reordering of packets that arrive out of order, etc. IP makes its best effort and TCP adds reliability. What Starlink needs first of all is a network protocol, at the level of IP, and the routing piece is likely to need to be smarter than IP's to handle Starlink routing peculiarities efficiently. QUIC is an experimental transport layer protocol so far supported by the Chrome web browser, and experimentally in Edge, Firefox, and Safari--and not much else. TCP is used by almost the whole internet, not just web browsers. The TCP part of the TCP/IP stack isn't going away until just about every networking application on the internet is rewritten to use a different protocol. But networks can support multiple simultaneous Transport layer protocols over the Network layer. No doubt QUIC and TCP will both go over the Starlink network protocol, whatever it ends up being.
Quote from: DaveH62 on 09/07/2020 06:07 pmGreat question about the routing. Will they create a new routing protocol. They will need unique routing protocols. I did a short search, but didn’t scour Reddit. I see early research and a pilot for IRIS from Cisco. Early versions don’t appear to be scalable to meet the Starlink architecture. This will be one of and potentially the largest commercial network in the world. Once the “starlink laser cloud” is running it would have a significant advantage for high frequency traders hedging between markets, and all cloud service providers. Bezos os in a losing position, but will probably spends 10’s of billions to try to be a close competitor to support AWS. SpaceX appears to have a 3 to 5+ year advantage on any competition. Figuring out the logic to make this work well will be up there with the pioneers of the internet.The software and protocols already exist. Starlink just is a more extreme case which may require some tweaking of existing algorithms for routing. But also may not. With Cell phone data usage being a large part of the internet traffic these days the routing algorithms have moved in the direction of Starlink's case and may already be able to handle very well any Starlink routing problem.
Quote from: gongora on 09/07/2020 05:03 pmI've been wondering how the inter-satellite links would work. With optical links, don't the send and receive elements both need to be actively pointed at each other? If so then a satellite sending information can't randomly pick another satellite on some instantaneously calculated optimal path to send the data to, because that other satellite wouldn't know to receive the information. Wouldn't the data need to go over predefined routes within the constellation? With a large number of satellites there could be a large number of defined routes, but near the end points there may need to be some sub-optimal routing for a small number of hops. For really high value routes (New York to London or various other combinations of commercial centers) they could define routes to always keep those as low latency as possible. I always thought that configuring that mess would be way harder than building the hardware. Long term links with the two neighbors in your orbit and to others in the next plane over with the same altitude and inclination might be simple, but trying to figure routing and setting up links between sats in different altitudes and inclinations just hurts my brain.
I've been wondering how the inter-satellite links would work. With optical links, don't the send and receive elements both need to be actively pointed at each other? If so then a satellite sending information can't randomly pick another satellite on some instantaneously calculated optimal path to send the data to, because that other satellite wouldn't know to receive the information. Wouldn't the data need to go over predefined routes within the constellation? With a large number of satellites there could be a large number of defined routes, but near the end points there may need to be some sub-optimal routing for a small number of hops. For really high value routes (New York to London or various other combinations of commercial centers) they could define routes to always keep those as low latency as possible.
Each Starlink satellite incorporates 60-69 (exact count unclear) modern(ish) Intel compute cores on the same fault-tolerant computer boards used for Dragon and Falcon 9. To the extent that we are willing to ascribe any good faith to Tim Farrar's arguments, perhaps he's not taking into account the unprecedented processing power of these satellites to run software routing algorithms. These aren't your father's rad-hardened satellite computers, and continuous iteration / continuous deployment of high-performance software systems is one of the things that every Elon Musk company does best. Starlink is almost certainly designed on the premise that they are going to be pushing frequent software updates to the satellites and the terminals to keep improving the network as the constellation is built out and the userbase grows. That's how Elon likes to engineer his products.
These satellites have extremely high bandwidth connections to the ground. There's no reason the heavier duty routing calculations couldn't be run on the ground.
André Staltz@andrestaltz·Feb 25, 2018If SpaceX's Starlink will provide internet access, obviously it will also provide a single IPv6 network, and could be a great opportunity for NAT-less peer-to-peer connections within Starlink only. I hope @elonmusk chooses wisely.Elon Musk@elonmusk·Feb 25, 2018Will be simpler than IPv6 and have tiny packet overhead. Definitely peer-to-peer.
Quote from: gongora on 09/07/2020 05:03 pmI've been wondering how the inter-satellite links would work. With optical links, don't the send and receive elements both need to be actively pointed at each other? If so then a satellite sending information can't randomly pick another satellite on some instantaneously calculated optimal path to send the data to, because that other satellite wouldn't know to receive the information. Wouldn't the data need to go over predefined routes within the constellation? With a large number of satellites there could be a large number of defined routes, but near the end points there may need to be some sub-optimal routing for a small number of hops. For really high value routes (New York to London or various other combinations of commercial centers) they could define routes to always keep those as low latency as possible.I was thinking control vs data... a radio link sends a "I want to connect" message, and the ACK means "ready to receive, my laser receiver is pointed your way, send it!!!"(this is a different question than deciding which bird to talk to)
Quote from: Lar on 09/08/2020 12:52 amQuote from: gongora on 09/07/2020 05:03 pmI've been wondering how the inter-satellite links would work. With optical links, don't the send and receive elements both need to be actively pointed at each other? If so then a satellite sending information can't randomly pick another satellite on some instantaneously calculated optimal path to send the data to, because that other satellite wouldn't know to receive the information. Wouldn't the data need to go over predefined routes within the constellation? With a large number of satellites there could be a large number of defined routes, but near the end points there may need to be some sub-optimal routing for a small number of hops. For really high value routes (New York to London or various other combinations of commercial centers) they could define routes to always keep those as low latency as possible.I was thinking control vs data... a radio link sends a "I want to connect" message, and the ACK means "ready to receive, my laser receiver is pointed your way, send it!!!"(this is a different question than deciding which bird to talk to) The best thing I can come up with is everybody monitoring a master control channel. Sort of like a trunking system times several thousand. The hardest part would coming up with cool names for the controllers. Several of them rotating duty and picking up if something goes wrong. Do an upgrade and it goes south, you just turn it off and the next one in line picks up using the previous version.
Quote from: geza on 09/07/2020 10:31 amFirst orbital test of optical com between Starlink sats.https://www.teslarati.com/spacex-starlink-space-lasers-first-orbital-test/This is significant.Just need SpaceX direct confirmation.
First orbital test of optical com between Starlink sats.https://www.teslarati.com/spacex-starlink-space-lasers-first-orbital-test/
Quote from: gongora on 09/07/2020 05:03 pmI've been wondering how the inter-satellite links would work. With optical links, don't the send and receive elements both need to be actively pointed at each other? If so then a satellite sending information can't randomly pick another satellite on some instantaneously calculated optimal path to send the data to, because that other satellite wouldn't know to receive the information. Wouldn't the data need to go over predefined routes within the constellation? With a large number of satellites there could be a large number of defined routes, but near the end points there may need to be some sub-optimal routing for a small number of hops. For really high value routes (New York to London or various other combinations of commercial centers) they could define routes to always keep those as low latency as possible.My understanding/speculation:Each sat has 4 laser connections, I assume 2 for the in-plane and 2 for out plane connections. These lasers can track satellites individually, within reason of the pointing of the host satellite.Sats in the same orbital plane: Here the host sat could potentially switch to further away satellites by slightly pointing the laser to a different sat. But realistically, say each sat is connected at all times to its predecessor and successor in the orbital plane. All sats know where they are relative to each other by communicating with ground stations. Sats out of plane: The laser trackers have to do a lot of work pointing to the correct sats. Sats that are on neighboring planes move relative to each other. Not sure how this works, but there is probably a fixed rule how the sats are connected and they hand off connections pretty fast.
Quote from: Semmel on 09/08/2020 07:10 amQuote from: gongora on 09/07/2020 05:03 pmI've been wondering how the inter-satellite links would work. With optical links, don't the send and receive elements both need to be actively pointed at each other? If so then a satellite sending information can't randomly pick another satellite on some instantaneously calculated optimal path to send the data to, because that other satellite wouldn't know to receive the information. Wouldn't the data need to go over predefined routes within the constellation? With a large number of satellites there could be a large number of defined routes, but near the end points there may need to be some sub-optimal routing for a small number of hops. For really high value routes (New York to London or various other combinations of commercial centers) they could define routes to always keep those as low latency as possible.My understanding/speculation:Each sat has 4 laser connections, I assume 2 for the in-plane and 2 for out plane connections. These lasers can track satellites individually, within reason of the pointing of the host satellite.Sats in the same orbital plane: Here the host sat could potentially switch to further away satellites by slightly pointing the laser to a different sat. But realistically, say each sat is connected at all times to its predecessor and successor in the orbital plane. All sats know where they are relative to each other by communicating with ground stations. Sats out of plane: The laser trackers have to do a lot of work pointing to the correct sats. Sats that are on neighboring planes move relative to each other. Not sure how this works, but there is probably a fixed rule how the sats are connected and they hand off connections pretty fast.If you watch the video in @NaN's message #66 above, video pos 1m50s, you can see the links are rather stable satellite to satellite wise from one plane to the neighboring planes. The satellites all kinda move together. This means satellite to satellite links in the network are long lived. The only issue is the sides swap twice an orbit.
