https://doodle.com/poll/avrcb7thiyawtvug
http://docs.automotivelinux.org/docs/apis_services/en/dev/reference/signaling/architecture.html#query-and-filtering-language & http://docs.automotivelinux.org/docs/apis_services/en/dev/reference/signaling/high-viwi-architecture.html
Laurent: how we start off with
Fulup's impression and how he reached it, then we can figure
out our documentation
... I have seen this happen before
Gunnar: I believe we are also looking at a document that is over a year old and his view may have changed
Fulup: CAN acquisition for us has
been driven by our underlying application framework at AGL, it
defines a security model
... we were motivated to move away message broker for our
security model
... sockets and rest were the same approach for audio
... we differed by starting from lower in the architecture
whereas you started from a high level api
... as we were not willing to reinvent. we chose OpenXC
provided by Ford
... it was not designed to run on linux but a controller
... we took various existing technology as input. a year ago we
only had a low level api for CAN
... in selecting a solution we chose based on what aligned with
our architecture better
... we found ViWi a closer fit
... we implemented a signal composer so you can send custom
notifications
... we are finding developers readily understand ViWi
... it took us only about two weeks to implement it on
AGL
... we didn't implement everything but most of it
... we do not have a clear understanding from OEM in what they
want from an app ecosystem
... we received a presentation recently in Japan on signal
composer which is different from ViWi and they are fine with
it
... we still do not have entirely what a developer would want
defined
... I am not sure how much you are looking at efficiency and
performance
... DensoTen liked the model and concept but were not willing to
give 10% of cpu for composing signals
... on gen3 3% of cpu is acceptable. we know there are
improvements to make, too many lookups and not enough
hashtables for instance
... we have been focusing more on performance and usability
than the semantics
... we are happy to deploy any method that would be used by
developers
Gunnar: VSS was a choice back when you chose OPENXC
Fulup: it wasn't doing the decoding yet of the underlying binary messages because that is what we get with openxc
Gunnar: Signal decoding details using CANdb as a field in the VSS database was presented at that time.
VSM (Vehicle Signal Manager) project was also announced before the need for Signal Composer that you described.
Fulup: I was at that
presentation, at the time it was written in Python
... ViWi can be distributed across multiple servers
Gunnar: VISS or any other
proposal should be possible to spread over multiple
servers
... your colleague misunderstood the specs about that
possibility, we discussed and agree that should be supported if
desired
... very interested to hear your input on optimization or how
to bind to lower level parts
... you said you already had a REST like thinking about how you
wanted access, hence ViWi would have made more sense
Fulup: the framework already had
a sense of verbs for granting priviledges to different
areas
... we wanted to leverage as much as possible in our existing
architecture
... for monitoring we have a completely different semantic and
using graphml(?)
Gunnar: can you elaborate on the
move from message broker? it was never a popular choice in
Genivi and came from the Tizen days
... the paper dwelled on this implementation detail
Fulup: that perhaps came from
Denso
... we could have fixed the issues but not with the
cybersecurity model at AGL
Gunnar: you implied we wanted to use automotive message broker at Genivi and we did not
Fulup: the reason we moved to a new model was because of the application framework
Laurent: interested to hear VISS
can be run on multiple servers, that is new to me as well
... I want to know more about decentralized VISS explored
further and in more detail. I found this a drawback
Gunnar: why wouldn't it be?
Laurent: the query model
Gunnar: same applies to a REST call to a URI not on a given server
Laurent: with a central tree approach, it suggests singular server
Fulup: the composed signal is
produced somewhere else and should carry the priviledge down to
the component that delivers the signal
... outside of performance that is quite a complexity
Laurent: there might still be
some things that are not clear to some of us
... still not clear how the single tree of VSS would be spread
across servers, ViWi is distributed but with a registry
... I am not aware of an implementation that does so
Ted: one thing I have been hearing
from Tier 2 is that they are considering whether or not they should be
implementing VISS on underlying ECU. that might be facilitated
with ethernet as OEM are considering alternatives to CAN and other
buses.
then a VISS instance can simply aggregate the
underlying ECU services, enforce authentication and access
I
am wondering if you are seeing others considering such a solution
Fulup: yes and no, most OEM have
gateways where they collect from different networks, firewalls
etc
... in case of AGL we support different network backends
... today the biggest demand is CAN and LIN since people want a
cheaper bus, agree ethernet is desireable
... that is independent of the server
... today in AGL you have to tell the signal composer where the
signal is. there is no discovery possible
Gunnar: I proposed before vehicle
signals to be a different domain and to be optimized explicitly
for them and separate from media and other apis
... I believe we will gain more from optimizing solution for
vehicle signals separately instead of trying to fit everything
into the same model
... I am not against a common solution but not at the cost of
performance
... database based browsing of media is different
Fulup: I agree. what people are
willing to do is merge signals for performance analysis
... today we are pushing most of the information into a
database to do some deep learning and to do so need to merge
everything
... vehicle signals are clearly different than a database
<Zakim> ted, you wanted to ask about developer survey
Fulup: low level CAN is being
used by AWS for instance
... Toyota had a demo running recently
... signal composer no, that is more at PoC and research
... we learned of the performance issues from Denso
Ted: we have some customers of this service (developers) in the conversation but can certainly use perspective from more. do you have a developer community using your signal composer?
Fulup: there is ISO work on doing
a common model for the cloud
... presently people are simply going with lower level CAN
because that is what they have
... we are lacking clear guidance on what developer community
would like to have
... we have good lower level adoption but we still need to
advance something for higher level
Ted: we might want to come up with something better for distributed services so apps do not have to be customized for a given vehicle, discovery perhaps
Gunnar: that is close to the fixed versus dynamic address, discoverability
[adjourned]