Saksham Agarwal 1 ti Recap: what is a computer network? A set of network elements connected together, that implements a set of protocols for the purpose of sharing resources at the end hosts
• Three important components
• Network elements • Core “infrastructure” • Protocols • Needed to use the network • Purpose • Sharing resources at end hosts (i.e. compu ng devices) ti Recap: what is a computer network? A set of network elements connected together, that implements a set of protocols for the purpose of sharing resources at the end hosts
Switch Router
Link
End hosts Recap: what do computer networks do? A computer network delivers data between the end hosts
• One and only one task: delivering the data
• Data delivery is enabled by
• Chopping the data into packets • Sending individual packets across the network • Reconstruc ng data from packets at the end hosts ti Let’s use a simpler representa on for the network in this lecture
ti Recap: what is a computer network? A set of network elements connected together, that implements a set of protocols for the purpose of sharing resources at the end hosts A computer network can be abstractly represented as a graph
Source Path
Source
Des na on Des na on ti ti ti ti Sharing the network (graph)
Source Path
Source
Des na on Des na on ti ti ti ti The problem of sharing networks • Must support many “users” and/or “applica ons” at the same me
• Each user/applica on wants the use the network (send and receive data)
• Limited resources • Network has di erent resources (as we’ll see over the course of this semester)
• Fundamental ques on
• How does the network decide which resource to allocate to which user/applica on at any given point of me? ti ff ti ti ti ti ti Resources relate to performance What are the performance metrics? Performance metrics in computer networks! • Bandwidth: Number of bits sent per second (bits per second, or bps) • Depends on • Hardware Bandwidth • Network tra c condi ons •… Delay
• Delay: Time for all bits to go from source to des na on (seconds)
• Depends on • Hardware • Distance • Tra c from other sources •… • Many other performance metrics (reliability, etc.)
• We will come back to other metrics later…
ffi ffi ti ti ti What are the various mechanisms for sharing networks? Two approaches to sharing networks • Reserva ons
• On demand ti Two approaches to sharing networks • First: Reserva ons • Reserve bandwidth needed in advance • Set up “circuits” and send data over that circuit • Must reserve for peak bandwidth
• How much bandwidth to reserve?
• Applica ons may generate data at rate varying over me • 100MB in rst second • 10MB in second second… • Reserva ons must be made for “peak” ti ti fi ti ti Circuit switching: Implemen ng reserva ons since… • Telephone networks (since the 19th century!)
• Mechanism • Source sends a reserva on request for peak demand to des na on • Switches/routers establish a “circuit” • Source sends data • Source sends a “teardown circuit” message ti ti ti ti ti Circuit switching: an example (red request fails)
Request = 100Mbps Source
Link Bandwidth = 100Mbps
Request = 100Mbps
Source
Des na on Des na on ti ti ti ti Circuit switching: an example (red request succeeds)
Request = 10Mbps Source
Link Bandwidth = 100Mbps
Request = 10Mbps
Source
Des na on Des na on ti ti ti ti Circuit switching and failures • Circuit is established
• Link fails along path (!!!!)
• First me we have seen failures making our life complicated • Remember this moment • It’s gonna happen, over and over again
• Must establish new circuit
• Circuit switching doesn’t route around failures!
ti Circuit switching summary • Goods • Predictable performance • Reliable delivery (assuming no failures)
• Not-so-goods • Resource underu liza on • Blocked connec ons • Connec on set up overheads • Per-connec on state in switches (scalability problem) ti ti ti ti ti Two apparoaches to sharing networks • Second: On demand (also known as “best e ort”) • Break data into packets • Send packets when you have them • Hope for the best
ff Packet switching: an example
Source
Link Bandwidth = 100Mbps
Source
Des na on Des na on ti ti ti ti Packet switching: what does a switch look like?
Input Ports Output Ports
Packet Bu ers ff Packets • Packets carry data (are bag of bits): • Header: meaningful to network • Body: meaningful only to applica on • More discussion in next lecture
• Body can be bits in le, image, whatever
• Can have its own applica on “header”
• What informa on goes in header?
ti fi ti ti What must headers contain to enable network func onality? • Packets must describe where it should be sent • Requires an address for the des na on host
• Packets must describe where its coming from
• Why? • Acknowledgements, etc
• That’s the only way a router/switch can know what to do with the packet ti ti ti Packet switching summary • Goods • No resource underu liza on • A source can send more if others don’t use resources • No blocked connec on problem • No per-connec on state • No set-up cost
• Not-so-goods • Unpredictable bandwidth availability • Unpredictable delay/latency • Packet header overhead ti ti ti ti Circuits vs packets • Pros for circuits • Be er applica on performance (reserved bandwidth) • More predictable and understandable (w/o failures)
• Pros for packets
• Be er resource u liza on • Easier recovery from failures • Faster startup to rst packet delivered tt tt ti fi ti ti Summary of network sharing Sta s cal mul plexing • Sta s cal mul plexing: combining demands to share resources e ciently
• Long history in computer science
• Processes on an OS (vs every process has own core) • Cloud compu ng (vs every one has own datacenter)
• Based on the premise that
• Peak of aggregate load is << aggregate of peak load
• Therefore, it is be er to share resources than to strictly par on them…
ti ti ti ti ti ti tt ti ti ti ffi Two approaches to sharing networks • Both embody sta s cal mul plexing
• Reserva on: sharing at connec on level
• Resources shared between connec ons currently in system • Reserve the peak demand for a ow
• On-demand: sharing at packet level
• Resources shared between packets currently in system • Resources given out on packet-by-packet basis • No reserva on of resources ti ti ti ti ti ti fl ti Internet Design Goals David Clark • Wrote a paper in 1988 that tried to capture why the Internet turned out the way it did
• It describes an ordered list of priori es that informed the design decisions
• What do you think those priori es were?
ti ti Internet Design Goals • Fundamental goal (#1): connec vity between exis ng interconnected networks
• Second level goals:
Packet Packet switched network switched network • #2 Survivability in the face of failure • #3 Accommodability to variety of services Unifying “network” protocol with minimal set of assump Packet switched network ons • #4 Accommodability to variety of networks + Support for mul ple • Other goals: “transport” protocols • #5 Ability for distributed managenement of resources • #6 Cost-e ec veness of the architecture • #7 Ease of a aching a new host to the network • #8 Accountability of the resources ff tt ti ti ti ti ti Something to think about… • What goals are missing from this list? • Sugges ons?
