Hypernet Overview

Transcript

1. Hypernet INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP jklsemi.com February 2014
2. Hypernet • Provide a p2p PaaS that: - makes computing cheap - helps fund the internet - gives users access to more free content, software and services over a private, secure and more reliable internet INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
3. Snapshot • Deploying an infrastructure product on March 15 ! • Selling distributed infrastructure at dramatically reduced prices (compared to the lowest cost providers) ! • VPE of SoundCloud has joined our advisory board. ! • Currently vetting developers for inclusion of 3-5 newer applications in our beta program INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
4. Snapshot • Product provides persistent storage of lots of data along with a broadcast system to notify peers of changes to storage. ! • Initial use cases are as follows: • Web crawling and data processing • Video and image encoding • Scientific research INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
5. Snapshot • We are building our distributed network on a peer-to- peer framework. ! • Application developers who build on our SDK will receive payment in exchange for the compute resources they contribute. ! • Currently vetting application developers for inclusion of up to 20 partners in our supply beta INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
6. AWS Akamai iCloud Rackspace Google Facebook Softlayer 0.5 1.0 1.0 1.3 1.3 1.4 3.8 The Market Hypernet Addresses In selling compute resources, Hypernet aims to disrupt the PaaS, IaaS and infrastructural software spaces, which represent a $100B market. INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
7. Hypernet Business Model INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP compute resources compute resources compute resources money money free software, services & content Internet User Application, service or content provider Hypernet p2p PaaS Cost-sensitive, compute hungry org
8. The Hypernet p2p PaaS Hypernet offers customers compute resources at dramatically reduced prices: • Storage • Bandwidth • Computing - Map/Reduce - Agents/Blackboards - General purpose computing INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
9. Types of Customers Buying Compute Resources Functionality Ideal Customer Type Q2 ‘14 Map/Reduce Q3 ‘14 File Sharing Q4 ‘14 CDN 2015 Full featured web applications INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
10. Pricing • Customers pay only for CPU cycles used and never pay for idle reserved instances INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP Transfer GB/month $0.10 $0.01 Storage TB/month $90.00 $9.00 Hypernet Industry TFLOP/hour $2.00 $0.20 • Hypernet is 10x cheaper compared to leading PaaS providers
11. Types of Partners Selling Compute Resources The focus for next 12 months is on new and existing application developers and content publishers with active user bases. Partners Distributors Apps & Aggregators Content Publishers Browsers OEMs INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
12. The Hypernet Monetization SDK • App developers and content publishers get paid per user per month: Rate Card device/month Smart Phone ~$4 Tablet ~$5 Laptop ~$7 Desktop ~$27 • One line of code to integrate • Static library for native desktop and mobile apps • Browser plugin for web apps INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
13. End User Value Prop INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP Users get content, services and software for free.
14. The Hypernet Technology Stack Networking Storage Map/Reduce Agents/Blackboards Services • Networking and storage are complete ! • Map/Reduce is expected in Q2 ! • Agents/Blackboards have been prototyped to create a proof of concept distributed CRUD web-app INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
15. Team Hypernet’s engineering team has a combined 45 years experience in software engineering and more than 20 years of experience building systems, p2p networks and internet-scale applications as well as the design and implementation of high-level computer languages. ! Hypernet’s business team has a combined 15 years of experience in enterprise-technology sales and a decade of experience in pro services, digital advertising and monetization, publisher yield management and application go-to-market. ! Vlad Tzankov (research and architecture) Founder of Inray, which developed machine learning, signal processing and p2p systems for Publicis, Air France, JP Morgan Chase and Casio Japan. Also led audio fingerprinting efforts at SoundCloud. ! Krasi Marinov (engineering and tech team) Led machine learning and networking engineering at Instinctiv, and then SoundCloud. ! Ant Taylor (biz dev, sales, mktg and biz team) Ant was an early product executive at Right Media (acquired by Yahoo) and ad cofounder at Moat. ! Peter Brodsky (product and engineering/business strategy) Background in AI and founder of two previous startups, including Instinctiv which licensed recommendation, DSP and p2p technology to Nokia, Yahoo and Amazon before being acquired by SoundCloud. INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
16. Recap • High-margin billion dollar opportunity • Cheap compute resources for those who need them • Payouts to companies with audiences that have compute resources • Private, secure internet with more free stuff for the end user • Team that has bootstrapped to deliver a prerelease of the p2p OS INTRO DEMAND SUPPLY END-USER TECH BUDGET TEAM RECAP
17. Q&A
18. Do end users have enough compute power to replace servers and aren’t mobile devices especially weak? Mobile CPU v Desktop CPU GFLOPS 0 0.1 0.2 0.3 0.4 2011 2011.5 2012 2012.5 2013 2014.5 World wide usable GFLOPS 3% 59% 12% 27% smart phone tablet PC server
19. What percentage of end users’ unused compute resources is actually harvestable? 0.0 0.2 0.5 0.7 0.9 mobile laptop desktop CPU Bandwidth Storage
20. What about latency? • Round trip ping times are usually faster because the end points are closer • Queries don’t wait in a queue because load is spread very thinly over lots of nodes • “Latency” that’s a function of waiting for data to transfer is also often faster as a result of peered downloads
21. What about bandwidth? • Bandwidth is often better as a result of peered downloads — consider that bittorrent is often faster than youtube • The storage and compute layers work to avoid moving data around the network in order to service queries
22. What are some of the networking stack components? • Anonymous routing • Fast routing • Peered downloads • Encryption • Sharded transmission • Offline delivery • Node uptime predictor
23. How does the storage layer work? • Provider independent security • Fast resource locator (~50ms) • Self-healing replication • Auto load balancing • SLA of 99.997% availability • Automatic grouping of related data on nearby machines
24. Aren’t user devices unreliable? Yes they are. So are servers. The Hypernet provides superior fault tolerance and SLAs through massively parallel redundancy.
25. What does a supply-side partnership look like? Native (digital) Hypernet (10%) ARPU $8.9 $99.0 Users 41M 4.1M Rev ($M) $360.0 $405.9 NYTimes model forecasts a 10x increase in ARPU
26. Could current applications run on this platform? Yes but portions of the code base would have to be migrated to the Hypernet platform. By way of example, migrating IMAP and SMTP required a week of work. The effort involved ripping out the networking layer and replacing it with calls to the Hypernet API.
27. How does Hypernet serve the users’ interests? • Paying for an application through Hypernet does not divulge a user’s identity • Applications built on the Hypernet PaaS are anonymous and secure • Users own their data • Despite sane defaults, users can also manually choose whom to compensate with their unused resources
28. Would users be comfortable sharing their compute resources?
29. Won’t harvesting users’ compute resources negatively affect the UX? Hypernet only harvests compute resources when the device is: ! ✔ Idle ✔ Plugged-In ✔ Fully-Charged ✔ On Wifi