CS514 Intermediate Course in Operating Systems Professor Ken Birman Vivek Vishnumurthy TA Putting it all together Today is our last lecture Wednesday was originally used as an in class final by Professor Schneider but we don t have an exam this year People interested in doing an early demo are encouraged to do so Wednesday or any time in the next two weeks All group members must be there Today s topic Synthesis Let s look back over the semester What s the big picture to take away Where will complex systems of systems go next What kinds of bets on the future are starting to emerge right now Technology adoption curve Market penetration Market Research saturationtopoint cross the chasm Technology is a hit quickly expands to fill the main niche Early adopters Cautious early are often Horror stories consensus lunatic fringe scare off community the mainstream Time The world we live in We re seeing Web 1 0 reaching that saturation situation For desktop uses the web is probably doing much of what it will do For wireless and mobile of course the situation is very different And we re using Web to mean web sites with relatively static content The world we live in Meanwhile Web 2 0 is taking off Technologies that leverage and support social networking Google mashups RSS feeds search Arguably Web 2 0 is already hitting its own saturation point The world we live in Web Services Basically can recognize these in terms of a set of simplistic steps Let s allow programs to do what browsers do Let s use Web Services standards to build systems of systems Let s make it easier to construct these solutions and interconnect them Call this a Web 2 0 technology area The world we live in Web 3 0 Makes for a fun homework topic someday you ll thank us But really just a distant glimmer right now The real Second Life system is just your basic datacenter very much a Web 2 0 construct Technology to support social networks Technology adoption curve Research to cross the chasm Market penetration The next new thing often occurs while the previous new thing is reaching stabilization Time A multi layered picture Over time a technology area such as web services ends up having wave after wave of major technologies Each follows a similar curve Assumes that there is a larger and larger aggregate market to pursue CS514 emphasis was on reliability mostly via replication We looked superficially at the technology backdrop against which all this is happening Client server interaction models CORBA we skipped this epoch Web Services the current new thing Systems of systems SoS of SOAs CS514 emphasis was on reliability mostly via replication Gossip technologies Very scalable and robust at least in some ways Predictable low load But sluggish poor choice if we want snappy response Other P2P technologies BitTorrent RON DHTs Some combine P2P ideas with gossip CS514 emphasis was on reliability mostly via replication Group communication Multicast but normally in support of replication or event notification Many types which leads towards a perspective that multicast type is a type much like any other type Object oriented multicast would probably look like live distributed objects Multicast type extends the component type CS514 emphasis was on reliability mostly via replication Against this backdrop we looked at What can and cannot be done FLP Scalable best effort multicast QSM Virtual synchrony model Consensus Paxos model Quorums and static membership Transactional replication 1SR Time critical and real time multicast Can view all of these as types of multicast and in fact QS 2 will do just this CS514 emphasis was on reliability mostly via replication Byzantine Agreement Strongest property of all Basically subsumes all the others Not impossibly slow anymore PRACTI BASE other BFT schemes But use only for sensitive purposes Giving rise to a vision Today Web Services focuses on how to connect clients to datacenters and more and more how to create complex SoS structures with datacenters that talk to one another But existing platforms offer relatively little autonomic support and forces us to build datacenters more or less entirely by hand The vision Systems that are Easy to build Better tools Autonomic by construction The tools lead us to robust solutions that can manage themselves in large complex deployments The tools themselves are better integrated into environments like NET Unlike cs513 we didn t look at security but even so add secure to this list Approaching that vision Cornell approach We need better technology Then show how it can integrate seamlessly into major platforms Then hope the world will imitate us The problem The world is drowning in a sea of noise technologies buzz Approaching that vision Corporate players Google is driven mostly by search and social networking opportunities Which for them are opportunities to leverage their role by helping you find their partner s sales sites or posting just the right ad at the right moment Many betting that Google is dead on Approaching that vision What about Microsoft MSN Live intended to enter same space But unclear so far just what the Live concept will really do Could Live be Live distributed objects Cornell thinks so but MS hasn t shown much sign of believing this Yet big success of NET is its clean integration of components clean use of type system So where are we Market penetration Could MSN and Google already be nearing the Research to satuation cross point the chasm Time Betting that our time is up If we bet that the datacenter search paradigm is already close to its peak Microsoft s next bet is on systems of systems but the technology is full of holes LooksSoS much conceptlike that early adopter scenario we discussed Research to cross the chasm Market penetration Time Betting that our time is up Google is aimed at cell phones Building a national free network lure in the marks with a loss leader Faustian bargain Just agree to run Google on your cell phone Then they use GPS voice recognition etc to somehow get you into their hotels restaurants nightclubs stores Google s problem Cell phone screens are just too small Already need to squint to see anything on them And voice recognition doesn t work very well yet an A I challenge for decades with progress but rather slowly Will Google pull it off What about us We re the crowd that ends up dealing with today s challenges These are basically Building datacenters with inadequate tools Making systems self managed even though Web
View Full Document
Unlocking...