A Review of Evolving Network Technology Ethernet & IP With associated infrastructure. J.J. Ekstrom IT 529 Thursday, January 15, 2015

Who is winning? l Ethernet has won the LAN wars l Ethernet is winning the MAN wars –Utopia, iProvo, Comcast.. l Ethernet is contending for part of the WAN… PPOE (Point to Point over Ethernet) l IP has won all best-effort wars wars… –Most ATM traffic is IP –A large portion of Sonet Traffic is IP l IETF and Vendors making IP transport of choice –Voice over IP –IP Multicast Streaming l Pretty much everything new assumes Ethernet packets with IP in them.

Why? l Simple transports l Work faster and cheaper l Put the smarts where it can work for more transports l Not as much advantage to smarter transports

Ethernet Characteristics l Ethernet shared media cable l Cable access method (CSMA/CD) l Unreliable Packet Delivery l Assumes higher layers do most of the work l Simple and Relatively fast on whatever physical transport with any generation of hardware.

Ethernet Shared Media Cable 1 l Physics determined the maximum length of the Ethernet cable –signal strength –cable characteristics

Ethernet Shared Media Cable 2 l All stations (nodes) hook to, and share a single cable

Ethernet Shared Media Cable 3 l Each station “listens” as it transmits

Ethernet Shared Media Cable 4 l Each station must transmit a minimum of 64 bytes to “fill” the cable before it stops listening

Ethernet Shared Media Cable 5 l If a 2nd node transmits before the 1st node finishes, the two transmissions collide and they must retransmit

Ethernet Cable Access Method (CSMA/CD) l CSMA/CD is a media-access method used by Ethernet and networks l CSMA/CD stands for Carrier Sense, Multiple Access / Collision Detection

How CSMA/CD Works - 1 l A station wishing to transmit first listens for traffic on the cable indicated by a carrier signal (CSMA/CD-Carrier Sense)

How CSMA/CD Works - 2 l If the carrier signal is detected, the station waits a period of time and tries again

How CSMA/CD Works - 3 l If NO carrier signal is detected, the station starts transmitting its packet (min. of 64 bytes) and simultaneously listening

How CSMA/CD Works - 4 l TWO stations can start transmitting at the same time (CSMA/CD - Multiple Access)

How CSMA/CD Works - 5 l If this happens, both stations hear garbage (CSMA/CD - Collision Detection)

How CSMA/CD Works - 6 l When collisons are detected, both stations : –cancel transmissions by sending a jam signal –wait a random amount of time before trying to transmit again

PROBLEM #1 l Physics doesn’t allow you to have LAN wires as long as you would like.

SOLUTION #1 l Repeater extended wire length, broadcast domain, and collision domain

PROBLEM #2 l Too many collisions. LAN wouldn’t carry enough traffic.

SOLUTION #2 l Bridging segments extends broadcast domain without collisions: Bigger LANs

PROBLEM #3 l Broadcast storms - result from multi-port bridges “flooding” all ports when packet destination is unknown and a loop exists.

PROBLEM #3 – when the original packet returns to a previous bridge, new packets are generated and a “storm” is generated.

SOLUTION #3 l D (spanning tree) installed on bridges. l Routers

SOLUTION #3.1 l 802.1D (Spanning Tree) added to bridges. –Spanning Tree is an algorithm that runs on bridges to eliminate loops dynamically.

SOLUTION #3.2 l Routers - make every segment another network or subnet by refusing to pass through any packet whose address it does not recognize.

SOLUTION #3.2 l NOTE: –in XNS a single broadcast domain is called a “network.” –in TCP a single broadcast domain is called a “subnet.” –network personnel often call a collision domain a “segment.”

PROBLEM #4 l Topology and failure characteristics - problems with bus-oriented LANs (i.e., when the wire breaks NONE of the stations can communicate).

SOLUTION #4 l Twisted pair LANs. –When any one wire segment fails, the whole LAN does NOT go down.

PROBLEM #5 l Not enough Bandwidth –only 10 MBPS available on each collision domain

SOLUTION #5 l Switches (multiport Bridges) - allows more segments (bandwidth) at a lower cost per port.

PROBLEM #6 l Controlling User Connectivity –keep groups separate –easily share resources between groups –do adds, moves, and changes without rewiring

SOLUTION #6 l VLANs of various forms create isolated broadcast domains (networks) l Connection between Virtual LAN networks requires a router. l People do security in their routers and firewalls at network boundaries anyway

Problem #7 l During roughly the same year period Token-Ring LANs, FDDI, ATM, and several other LAN and WAN technologies have been undergoing similar evolutionary tracks as ethernet. l It was not clear that there would be a clear winner. l How do you hook them together and protect your technology investments? l Users don’t care how their bits get pushed around, only that things work.

Solution #7 l Internetworking…The real reason IP has won the protocol wars. –Works well on P2P links –Works well on LANs –Makes very few demands of participant networks –“Rough consensus and working code” l Motto of the IETF l The way to get useful things quickly in a world of confusion…what works best wins.

Internetworking: Internet, intranets Outline Best Effort Service Model Global Addressing Scheme

IP Internet l Concatenation of Networks l Protocol Stack R2 R1 H4 H5 H3 H2 H1 Network 2 (Ethernet) Network 1 (Ethernet) H6 Network 3 (FDDI) Network 4 (point-to-point) H7R3H8 R1 ETH FDDI IP ETH TCP R2 FDDI PPP IP R3 PPP ETH IP H1 IP ETH TCP H8

Service Model l Connectionless (datagram-based) l Best-effort delivery (unreliable service) –packets are lost –packets are delivered out of order –duplicate copies of a packet are delivered –packets can be delayed for a long time –(Sound like Ethernet?) l Datagram format VersionHLen TOSLength IdentFlagsOffset TTLProtocolChecksum SourceAddr DestinationAddr Options (variable) Pad (variable) Data

Problem: Different MTU l All LAN Technologies do not have same maximum packet size. l Network layer has no simple way to determine path l Packets dropped if too big to be forwarded

Solution: Fragmentation and Reassembly l Strategy –fragment when necessary (MTU < Datagram) –try to avoid fragmentation at source host –re-fragmentation is possible –fragments are self-contained datagrams –use CS-PDU (not cells) for ATM –delay reassembly until destination host –do not recover from lost fragments

Example Ident = xOffset = 0 Start of header 0 Rest of header 1400 data bytes Ident = xOffset = 0 Start of header 1 Rest of header 512 data bytes Ident = xOffset = 512 Start of header 1 Rest of header 512 data bytes Ident = xOffset = 1024 Start of header 0 Rest of header 376 data bytes

Problem: Global Routing l Next hop is always a local decision l How do you know which way to send a packet?

Global Addresses l Properties –globally unique –hierarchical: network + host l Dot Notation – – – NetworkHost A: NetworkHost B: NetworkHost C:

Datagram Forwarding l Strategy –every datagram contains destination’s address –if directly connected to destination network, then forward to host –if not directly connected to destination network, then forward to some router –forwarding table maps network number into next hop –each host has a default router –each router maintains a forwarding table l Example (R2) Network Number Next Hop 1 R3 2 R1 3 interface 1 4 interface 0

Problem: Network Address binding l Network Layer Address is logical and global l MAC addresses are bound to physical network l Point-to-Point may have no physical address

Solution: for IPX l Make network address include physical address l 16 bit Network number + 48 bit MAC address = 64 bit address

Solution: For IPv4 l Map IP addresses into physical addresses –destination host –next hop router l Techniques –encode physical address in host part of IP address l Assumes fixed host address l Doesn’t work with subnets or 48 bit MACs (IP is 32 bits) –table-based l ARP –table of IP to physical address bindings –broadcast request if IP address not in table –target machine responds with its physical address –table entries are discarded if not refreshed

ARP Details l Request Format –HardwareType: type of physical network (e.g., Ethernet) –ProtocolType: type of higher layer protocol (e.g., IP) –HLEN & PLEN: length of physical and protocol addresses –Operation: request or response –Source/Target-Physical/Protocol addresses l Notes –table entries timeout in about 10 minutes –update table with source when you are the target –update table if already have an entry –do not refresh table entries upon reference

ARP Packet Format TargetHardwareAddr (bytes 2–5) TargetProtocolAddr (bytes 0–3) SourceProtocolAddr (bytes 2–3) Hardware type = 1ProtocolType = 0x0800 SourceHardwareAddr (bytes 4–5) TargetHardwareAddr (bytes 0–1) SourceProtocolAddr (bytes 0–1) HLen = 48PLen = 32Operation SourceHardwareAddr (bytes 0–3)

Solution: IPv6 l Make Network Address 128 bits l Carry 64 bit IPX addresses l Carry 32 bit IP addresses l Even carry DEC Net and others l But big tables and smart routers!

Internet Control Message Protocol (ICMP) l Echo (ping) l Redirect (from router to source host) l Destination unreachable (protocol, port, or host) l TTL exceeded (so datagrams don’t cycle forever) l Checksum failed l Reassembly failed l Cannot fragment

Problem: Class based l (0)7 bit Class A too few networks, 6 million hosts too many l (10) 15 bit Class B still too few networks, 64,000 hosts still too many. l (110) 23 bit Class C still too few networks 256 hosts too many for many applications. l Address “ownership” companies grow, shrink, die …

Solution: Classless l CIDR – Classless Inter-Domain Routing l Block 20 bit network address Class ignored l 12 bit host = 4k hosts l ISP’s own blocks

Problem: Trust l ISP’s compete for carrier business l ISP’s want to give better service to their own customers l Typical routing algorithms require that routers trust all other routers l Rogue routers kill networks

Solution: Different Routing Algorithms l RIP – local routers trust each other l OSPF, IGRP, EIGRP– local trust with some security l BGP – Point-to-point manual configuration Router not obligated to use information. (How does the Internet ever work?)

Problem: Spanning Tree wastes bandwidth l Blocked links are not used. If they are 10 gig links that is a big deal. l Fail-over times were on the order of 1 minute. l Shutting down the entire spanning tree during recalculation is not acceptable.

Solutions: many small ones l Link aggregation allows redundancy and full use of the bandwidth except during failure. l Rapid Spanning tree allows much faster failover and doesn’t block everything while reconfiguring l Ports connected to end nodes don’t wait at all. (Portfast on cisco)

Problem: Latency in Hierarchy l Datacenters assume that each migration target has similar network performance to other VMs. l Traditional LAN topologies don’t guarantee this.

Solutions: Stir everything (SDN)