selectively active in a target service area. Finally, to maximize
the benefit from the LEO edge, which could play a role in
helping LEO satellites achieve economic feasibility, it will be
critical to enable capabilities for its on-demand, multi-tenant
and dynamic operation. In short, the LEO edge will require
orchestration capabilities, much like what is required (and
being developed) for the terrestrial edge [32, 37].
A good starting point in the design space for LEO edge or-
chestration are current terrestrial orchestration systems. How-
ever, the current design of the orchestration systems is hinged
on a tight coupling between the orchestrator and the infras-
tructure it manages, making it unsuitable for the LEO edge
due to its inherent characteristics. First, the time-of-sight of a
particular LEO satellite is severely restricted due to the con-
tinuous motion of the satellite with respect to the Earth. With
a satellite visible only for a period of
7 − 30
minutes depend-
ing on its altitude [
36
], it is imperative that applications be
rescheduled onto the next incoming satellite to prevent the
risk of reduced availability of the edge functions. Second, the
lack of accuracy in predicting the position of LEO satellites is
well known and continues to be an active area of research in
aerospace engineering [
24
,
38
]. This presents difficulties in
determining the position of, and thus in predicting the trajec-
tory of, the LEO satellites and further compounds the problem
of a limited time-of-sight.
Addressing the above-mentioned problems is already non-
trivial in terms of selecting the right LEO edge nodes and
determining a schedule for application orchestration. A so-
lution focused on these two aspects alone, would only be
sufficient for stateless applications like geographical image
analytics. For stateful applications, like a web cache, another
key aspect that needs to be handled is graceful state transfer.
If the state is rebuilt every
7 − 10
minutes, there would not be
much benefit that can be extracted from the edge functions on
LEO satellites. Without access to state, the idea of localiza-
tion, a key reason for edge benefits, cannot be realized in the
LEO edge. This highlights a clear need to improve the state
handling in orchestration to deliver tangible benefits from
LEO satellite edge.
In response, we present
Krios
1
– an orchestration system
for the LEO edge.
Krios
provides fundamental constructs and
directives which can be incorporated on top of any current
orchestration framework, with a few additional modules to
address the requirements of the LEO edge.
Krios
achieves its
goals through the key idea of loose coupling between edge
function orchestration and the underlying infrastructure.
Loose coupling in
Krios
is achieved through three new
mechanisms as part of its scheduling constructs. First, incor-
porating satellite path projection models in orchestration en-
ables
Krios
to predict the trajectory, and thus future positions,
of LEO satellites with respect to regions in which they are
visible (availability) with known bounds on the error, depend-
1
derived from the Greek god of constellations, Crius
ing on models used. This allows
Krios
to select appropriate
satellite nodes and (re)schedule the edge functions across the
LEO edge without impacting their availability in the target
region. Second, temporal compensation enables ahead-of-
time hand-off, i.e. deployment and preemptive killing of edge
functions on LEO satellites, masking the initialization times.
Third, cluster affinity chains that span individual clusters al-
low
Krios
to choose the satellite among many which may
be visible ahead of time, i.e. before those nodes join a given
ground station cluster. This reduces the time complexity of
choosing the satellite and makes deployment more predictable.
These mechanisms are also leveraged to orchestrate the edge
functions’ state transfers across the LEO edge.
In this paper, we use Kubernetes [
8
] to demonstrate the
benefits of
Krios
. It is, after all, the starting point for the ter-
restrial edge, as it leverages all the technology developments
in the datacenter space and has gained acceptance in industry
as the go-to orchestrator for web services in the cloud and
edge functions at the edge [7].
Summarizing, the contributions of this paper are: (i) quan-
tifying the limitations of terrestrial orchestration solutions
when used in the LEO edge due to the tight coupling to in-
frastructure inherent in their design, and (ii) design of an
end-to-end platform –
Krios
– that embodies the new system
support needed to loosen this coupling and to make LEO edge
orchestration feasible. Finally, we demonstrate the promise
of the approach through our preliminary experimental evalua-
tion.
2 Background
Case for Satellite Internet:
LEO satellites have been in exis-
tence since the late 1990s when they were designed to support
voice, data, facsimile, and paging. IRIDIUM was one of the
major players, having launched 95 satellites between 1997
and 2002. But LEO satellites could not compete with the
GSM services and hence, could not gain much traction. Since
then, a number of factors have changed, making a case for
the success of Internet services based on LEO satellites. The
biggest of those is the decreasing cost of space access, aided
by the commercialization of launch vehicles [
10
,
29
]. While
the lifecycle costs per satellite for Iridium were about US $ 5.7
billion [
3
], the estimated cost for a
4425
LEO constellation is
about US $10 billion [18].
In addition to this, while the first use case of LEO satellites
targeted the existing sets of mobile and Internet users, the
focus has now shifted to the next billion users. This provides
the benefit of manufacturing at scale and results in further
reduction of operational and manufacturing costs. Further,
there is a significant design difference in how the system was
supposed to work earlier in comparison to today. Earlier, a
user would need a specialized handheld device to communi-
cate with a satellite. These devices with significantly powerful
antennas (to directly connect to the satellites) led to higher