Credits to EKR and his minuting tool at ietfminutes.org
Session Date/Time: 06 Nov 2025, 14:30 UTC
This was the inaugural official meeting of the proposed SPACE (Systems and Protocol Aspects for Circumstellar Environments and Beyond) Research Group (RG) within the IRTF. The session began with an introduction to the RG's mandate, emphasizing its role as a research-focused entity rather than a standards-developing body, with a broad scope for exploring networking in space environments. Five technical presentations covered diverse topics including optical link scheduling, adaptive bitrate and congestion control for LEO networks, an AI-enabled testbed for satellite mega-constellations, measurements of LEO routing and addressing, and a proposed code for modeling satellite constellation topologies. Discussions highlighted the unique challenges of space networking, the need for robust simulation and testing tools, and a community-driven approach to defining future research directions.
The SPACE RG is a proposed IRTF research group, distinct from IETF standards bodies. Its focus is on conducting research related to networked nodes in space environments, characterized by high mobility, dynamic topologies, and environmental/resource constraints. The group's scope is intentionally broad, aiming to foster an open, exploratory community and avoid becoming narrowly focused on specific commercial systems (e.g., "just a Starlink measurement group").
Activities will include sharing research ideas, methodologies, software, and data to promote reproducibility and comparability. The RG will not develop protocol specifications, deferring this to relevant IETF working groups (e.g., DTN, TVR, TCPM, QUIC), but will provide insights that can inform their work.
The broad charter requires community input to identify specific areas of focus and initial work.
The agenda has five talks on different perspectives of satellite and space networking.
Optical inter-satellite links (OISLs) form an "Internet backbone in space," with satellites acting as routers. Advantages are that they allow higher data rates due to narrow beam widths. A major challenge is that extremely precise pointing is needed, with slow pointing, acquisition, and tracking (PAT) processes (tens of seconds for LEO, minutes for deep space). PAT delays necessitate scheduling links well in advance, rather than dynamic, real-time routing decisions based on congestion.
Various OISL types were discussed: in-plane, cross-plane, cross-seam, inter-shell, and space-to-ground (optical ground stations, OGS). Space-to-ground links are particularly challenging due to atmospheric issues and limited OGS availability, making efficient scheduling critical. Interplanetary networking amplifies these challenges, requiring constant selection from a minimal resource pool and focusing on sequencing, directionality, and maximizing contact windows.
Potential strategies for OSIL scheduling include minimizing retargeting delays, onboard data aggregation/buffering (leveraging DTN/Bundle Protocol concepts), and store-carry-forward. For LEO meshes, at large, no end-to-end disruptions are expected. Anything past the moon will be substantial and will require thinking differently about networking protocols.
Observations of Netflix streaming over LEO (predominantly Starlink) show good visual quality but frequent rebuffering and bitrate switches due to network variability.
Two specific areas are highlighted in the talk:
Congestion Control Experimentation
Modifying TCP New Reno to be "less reactive to loss" (multi-CP approach) increased congestion window and throughput on Starlink. This resulted in improved Quality of Experience (QoE) at lower quantiles (fewer rebuffers, stalls, bitrate switches). However, it led to a significant trade-off: over 300% increase in retransmissions and substantially higher tail latency.
Conclusion: Tweaking existing CC helps but can't fully overcome LEO's high variability (moving bottlenecks, non-congestive loss). Rate-based CCs like BBR and pacing show more promise.
Adaptive Bitrate (ABR) Experimentation
Tuning ABR parameters (throughput discount, low buffer threshold, throughput smoothing) can reduce rebuffering events. However, even with tuning, LEO performance (Starlink) does not fully close the gap with terrestrial ISPs, and quality trade-offs (e.g., reduced bitrates) persist. A need exists for "variance-aware" ABR algorithms that do not solely rely on single-point throughput estimates.
This leads to a "community Call" that emphasizes the need for realistic testing environments (open datasets, simulators, testbeds) and treating LEO as a first-class network environment in design, not an afterthought. The discussion confirmed that the "high variability" is considered inherent to LEO systems due to handovers and reconfigurations, not just Starlink specifics.
Discussions covered the QoE metrics, for which multipe metrics including video quality are leveraged. It is not possible (at the moment) to calculate the amount of bandwidth provided by Starlink to a particular country (e.g., through geofencing). The observed variability in link/path characteristics is not limited to Starlink; multo-layered LEO constellations could help here. SPACE RG could work on making protocols work for LEO constellations, similar to overcoming issues with protocol performance over GEO satellites before. Simulation tools that are deployment independent could be useful, backed by benchmarks and toolings.
Pablo presented an AI-enabled testbed for satellite mega-constellations being developed at Carleton University, motivated by the increasing complexity of these systems (thousands of satellites, OISLs, advanced on-board processing). The testbed simulates orbital dynamics (Matlab/STK), network protocol dynamics at the packet level (Omnet++), and integrates machine learning models.
It focuses on transport and network layers, abstracting link and physical layers. Includes satellites, user terminals, ground stations, and SDN controllers/switches. While the specific tool is not open-source, it leverages extensive open-source code (INET, NS-3, OpenFlow, H3 lib), suggesting potential for community replication.
Current research areas include QoS optimization (queuing, joint routing/queuing), SDN for dynamic controller deployment, feeder link management (RF vs. optical, weather forecasts), and green traffic engineering.
Shenping presenting findings on the current operation of the Starlink system based upon extensive measurements.
He discussed Starlink's specific IP addressing schemes and naming conventions for dishes/users (pop-specific) versus OneWeb's approach for enterprise users. Noted the lack of public naming for some Starlink backbone elements.
Concerning routing, Starlink dictates satellite connections, with regular 15-second, globally synchronized handovers causing latency spikes and packet loss. Faster switching is available for obstruction. The talk explored Starlink's global ISP-like ground backbone architecture, including identified Point-of-Presence (PoP) locations and their expansion. He also identified suboptimal Routing: Highlighted "home pop centric" routing that can lead to significantly longer paths than physically possible via OISLs (e.g., dish-to-dish traffic spanning continents). New PoPs help reduce this, but direct ISL routing is often more optimal.
In the context of PoP selection, he also pointed to a network/application Mismatch and llustrated trade-offs between PoP latency and content latency, particularly in regions where content caches might be geographically closer than the assigned PoP.
Finally, the talks introduced a cross-country LEO satellite testbed in Canada, emphasizing making all research (papers, datasets, code) publicly available due to difficulties in obtaining data directly from commercial operators.
Maxime presented his Internet Draft that proposes a standardized, non-ambiguous, and machine-parsable code for representing satellite constellation parameters (e.g., altitude, inclination, total satellites, orbits, phasing, Walker type). This code is based on Tim van der Lee's open-source format; it facilitates consistent description and computational analysis (e.g., inter-satellite distance calculations).
He acknowledged limitations in the initial draft (e.g., implicit fixed link configurations for mega-constellations) and suggested future work on link quality, constraints, and integration into a "space RG toolbox." Invited the community to try using the code and provide feedback.
SPACE RG will maintain a broad and inclusive scope, fostering diverse research, sharing methodologies, software, and data, and building an open, exploratory community.
The chairs will continue to foster community engagement and gather feedback to further refine the RG's focus areas, based on community interest and contributions. The group will seek to prioritize broad research topics over narrowly focused commercial system measurements.
The community is encouraged:
Participants should subscribe to the space@irtf.org mailing list and actively share ideas for future SPACE RG activities and topics.
The group plans to meet again co-located with future IETF events to continue technical discussions and advance research in space networking.