-
Notifications
You must be signed in to change notification settings - Fork 1
/
RE Protocol Labs RFP decision.txt
202 lines (192 loc) · 13.7 KB
/
RE Protocol Labs RFP decision.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
To: Jorge Soares; 'Yiannis Psaras'; David Dias
Cc: Jonathan Holt (jonathan@transendx.com); Jaime Llorca; 'ResNetLab'; 'Jonathan Gross'
Subject: RE: Protocol Labs RFP decision
From: Rhett sampson <rhett@gtsys.co>
Sent: Tuesday, 8 December 2020 12:12 PM
To: Jorge Soares <jorge@protocol.ai>; 'Yiannis Psaras' <yiannis@protocol.ai>; David Dias
<david@protocol.ai>
Cc: Jonathan Holt (jonathan@transendx.com) <jonathan@transendx.com>; Jaime Llorca
<jllorca@gmail.com>; 'ResNetLab' <ResNetLab@protocol.ai>; 'Jonathan Gross' <jpgross@protocol.ai>
Subject: RE: Protocol Labs RFP decision
Jorge, Yiannis, David and all
Apologies for the delay in replying to your email re our R&D bid. We’ve all been very busy in the back
half of 2020! Given our recent submission to the IPFS 2021
roadmap https://github.com/ipfs/roadmap/issues/79 , I thought a more detailed reply would be in
order. It’s long but will hopefully help you understand why we think there is such a good fit between our
technologies. Here goes:
This must be read in the context of some tectonic shifts in the tech industry and in the movie business
during 2020 that further “turns the screws” on TC/IPs ability to deliver:
a. Nvidia buying ARM (and Mellanox and Cumulus) changed everything. Overnight Nvidia
became not just the biggest network equipment vendor in the world, but bigger than all
the rest put together. And it put AI front and centre of the network business. The fit
with our SPAN-AI tech is perfect and we are already working with them on building a
prototype PoC on their DGX/EGX switches. See below.
b. Apart from breaking every network in the world, Covids changed the movie distribution
business forever. It broke the power of the cinema “blockbuster” chains. Disney went
100% online with Mulan. And Warners went day and date for its entire slate for 2021.
These are the two most powerful studios in Hollywood. The rest will follow as sure as
night follows day. It will increase load on the Internet significantly. See below.
To answer your specific points:
1. Intellectual property. Like you, to date we are an R&D organisation. You have been very
successful at securing funding to build the results of your research. We have not yet secured
funding to build the SPAN-AI prototype PoC. We do have network platform partners to
implement on: Nokia and Nvidia. We were hoping that Protocol Labs would provide part of that
funding and enable us to secure the balance. As you say, we know exactly what to build; and
how to build it. We have protected the “what” by patents to give us an equitable basis to
negotiate how we commercialise our IP. The “how” remains our trade secret for the moment. A
significant part of the “how” formed part of our second patent which has not yet been
published. In the interests of providing insight into that, we have released the architecture in
our discussion repo for merging SPAN-AI and IPFS. It was also sent to you as part of our response
to RFPs 7 & 8. We understand why it may appear confusing. It requires the benefit of many
months, years and decades of background in working out how to do this. We are happy to share
some of this with you, starting with your specific questions:
a. A general overview of how we combine NRR and NBR. (Note: we use “IPFS” to include
the whole protocol suite/stack).
i. Combining name resolution routing (NRR), e.g. TCP/IP and IPFS, with name
based routing (NBR) is EXACTLY what we are doing. As you say, they are very
different approaches. But it turns out they can be unified very simply. NRR, and
particularly the IPFS overlay of TCP/IP, is very good at content storage. As
everyone agrees, IPFS needs work to store large amounts of content,
particularly video, and in retrieving and distributing it. NBR is very good at
distribution and is part of the basis of our foundation patent. The two can be
combined by concatenating an IPFS hash address with a content name and
routing on that. There is a very simple mechanism to allow both TCP/IP and
name based routing, either in one combined switch/router or in single purpose
switch/routers. Beyond simple routing, machine learning AI adds intelligent
routing and intelligent distribution and storage of content in the network. We
call this combination intelligent Hybrid Adaptive Routing Design or AI-HARD.
This intelligent, non-deterministic routing is required to reach hyper scale.
Deterministic routing has reached its limits of growth. The NBR projects such as
CCN, ICN and particularly NDN have done an excellent job at laying the
foundations for this, particularly the stochastically balanced Pending Interest
Tables and Satisfied Interest Tables. AI-HARD combines both NRR and NBR for
storage and distribution respectively.
b. AI. AI is central to SPAN-AI-HARD. There are a number of component systems:
i. Local autonomous AI agent swarms. These, combined with a new multi-level
DHT and learned knowledge of demand and network state, provide optimised
local storage and distribution. The multi-level DHT comprises three layers:
topological, topic and combined. Topology includes local, regional and national
DHTs.
ii. Global optimising AI’s. These provide global predictive knowledge of users’
consumption/production patterns, application requirements, network
conditions (including overlay mesh health), and available resources for proactive
optimization.
iii. The operational health mesh, AmI-Rendevous, providing smart discovery,
configuration, and self-organization services.
iv. Jaime Llorca has been modelling next gen network optimisation at Bell Labs for
10 years. He says this is the only architecture that works.
v. Given wide adoption of SPAN-AI, there will be multiple global optimising AIs.
These will need to talk to each other to form a UCDN. This can be achieved by
standardising models and parameters across AIs. The global AIs are also
responsible for training and distributing agents and collecting mesh data and
can therefore enforce standards.
c. (Virtualisation). This was not a question but is critical. Containerization of SPAN-AI
agents enables a fully cloud switched network, conforming with definitions of virtual
network infrastructure. It also enables cloud testing of agent models at scale.
d. In-network storage. This is critical. It is a central part of the CCN/ICN/NDN architectures
and is one of the core components of SPAN-AI. The network becomes the cloud. Our use
of it is completely intentional and will be one of the core demonstrations of our
prototype PoC switches.
e. Dataplane-aware symmetric Interest-Data packet forwarding. Again this was
completely intentional. The Pending Interest Tables (PIT) and Satisfied Interest Tables
(SIT), along with optimising AI agents, form a separate control layer in the network that
manages storage and forwarding of content in the network “with SDN like capability”.
(everything old is new again!)
f. IPNS and DNSLink. Content addressing and routing at the network layer versus IPFS
overlay on TCP/IP. See above.
g. Multi-level DHT. See above. There are many algorithms to choose from, including non-
deterministic ones such as cuckoo hashing. We are modelling the optimum.
h. Storage centric versus distribution/delivery centric. See above.
i. In network load balancing. This is an inherent feature of both NDN and SPA. And very
nice to have.
j. Multicast. This is an inherent feature of both NDN and SPA. And very nice to have.
k. (Loop prevention). This is an inherent feature of both NDN and SPA. And very nice to
have.
l. Simulation. We are modelling in Matlab and using a state of the art optimisation tool.
There are many simulators to choose from, but perhaps the best is distributing
containerised agent algorithms in cloud hosting services such as AWS and testing at
scale. Unfortunately it is also very expensive.
2. Track record. You are right, we are a commercial organisation and have taken a different path
from you. That doesn’t mean our interests aren’t aligned. We just need to find a way to receive
a reasonable return for our investment and IP. You guys have managed to do that already very
successfully. We would love to work with you to find a path to that for us. Given the very limited
scope for response to the R&D RFPs, let me expand a bit on our track record now:
a. Intel and Microsoft (and Nvidia and Nokia and Bell Labs)
i. We were the first in the world to obtain a PlayReady 2 SL3000 (4K) security
certificate on an open Windows x86 platform. The platform was the 7th
generation Skull Canyon. This product was essentially our first Blust design
(www.blust.tv ), submitted to Intel about 12 months previously. To quote Kate
Burleigh, the CEO of Intel Australia at the time: “You aren’t going to sue us are
you Rhett?” My reply: “Do I look crazy Kate? We didn’t want to be in the
hardware business. We are more than happy to leave that to you”. Kate (now at
Amazon) will confirm this.
ii. We helped Intel close a security hole in the previous 6th gen NUC in order to do
that. It required working very closely with Intel and Microsoft to satisfy the
requirements of PlayReady on the NUC. The comment from William (Bill) Lewis,
principal engineer, Intel, at the time was “I wish we’d spoken to you guys 2
years ago”. Bill will confirm this. You will note he is a first level contact on my
LinkedIn. We also have copies of the confidential email exchanges between us,
Intel and Microsoft, sorting out who needed to do what.
iii. Intel understood the potential of Blust and SPA (our first protocol
implementation) as an open x86, next gen, p2p, content based routing platform
for the efficient delivery of movies. They were not able to fund us due to a $5M
turnover threshold for their VC investment group. Roland Wooster, at the time
Platform Innovation Manager, Desktop Platform Group, will confirm this. Roland
(now also a principal engineer at Intel for video applications) is also a first level
contact on my LinkedIn. Roland meets regularly with all major Hollywood
studios.
iv. Our second Blust product design was essentially a Skull Canyon NUC with the
world’s first mini-PCIE based Nvidia graphics chip. We worked extensively with
Gigabyte on the specs for this. It eventually became Gigabyte’s Brix gaming
platform. We worked with Michelle Huang at Gigabyte on this.
v. The Blust gaming design, with our SPA network tech, pre-empts the new Nvidia
DGX/EGX products by 5 years. Our design strategy and objectives are identical.
We even had an ARM design previously. We are implementing a prototype PoC
on Nvidia’s DGX and EGX platforms.
vi. Nokia understand the potential of SPA. We have a confidential Letter of Intent
from Nick Lochrin, Head of Commercial Business Development, Nokia
Enterprise. We are developing the first SPAN prototype on Nokia’s SR Linux data
centre switch platform.
vii. Nick introduced us to Jaime Llorca, then a senior research scientist at Bell Labs.
Jaime is now a research professor at NYU, an invaluable member of the GT
Systems team and a co-author of our SPAN-AI patent and our white paper.
Jaime is one of the leading network modelling and optimisation researchers in
the world https://scholar.google.com/citations?user=KSI2DE0AAAAJ&hl=en
viii. Jaime is modelling SPAN-AI right now. Based on 10 years of research at Bell Labs
into network optimisation, he is confident it will show significant improvements
in efficiency and QoS.
b. Hollywood and content retailers.
i. We have been working with every major Hollywood studio for over a decade, in
Australia and the US.
ii. In 2009 we brought manufacture on demand (MoD) of original CDs and DVDs in
retail stores to Australia. It was deployed in Harvey Norman and Big W. At an in-
store in Big W, the Australian CEOs of every Hollywood studio and game
publisher were present. This can be confirmed by Charlie Davey, then head of
home entertainment for Big W.
iii. We began R&D on our online delivery technology in 2010, at the direct request
of Big W with the support of the studios, led by Warner. Roger Clark, the then
CEO of Warner Australia, can confirm this. The classic statement came from
Aleisha McAulay, then head of digital at Fox: “The peer word makes us nervous,
but we’re coming with you”.
iv. We received technical approval from Sony for online distribution of their full
catalogue, including 4K, in 2016, from Tim Wright, then VP Worldwide New
Media and Technology. Tim is now Principal Technical Program Manager at
Amazon. In his approval email he said “Thanks Rhett and Andrej [our head of
digital security and ex head of security for Disney]. This [Blust] has technical
approval (subject to contractual agreement). Thanks very much for your
comprehensive response. If only all licensees were like this! [our emphasis]”
v. We are in progress with all the other studios for technical approval. The head of
digital security for Disney said to us at a meeting in Burbank: “You solve a huge
problem for us in Asia”.
vi. We have been in discussions with Paramount in Australia and LA regarding a
proof of concept. They were very interested in p2p distribution and crypto. This
can be confirmed by Anne Masuda, then senior manager of digital distribution
for Paramount, now VP of global SVOD distribution for ViacomCBS.
vii. Given recent developments in digital distribution, both Hollywood and
technology, we strongly believe we could gain participation by one or more
Hollywood studios in a next gen digital distribution PoC.
To quote Humphry Bogart at the end of Casablanca, we “think this is the beginning of a beautiful
friendship”. Looking forward to further interesting discussions.
Warmest
Rhett Sampson
<original email not included>