lecture 1 week 1 - murdoch universityftp.it.murdoch.edu.au/units/ict227/b527/lectures/b227… ·...

Congestion Prevention

B227 Data Communications Lecture 7-19

Peter Cole 2001

Flow ControlFlow control is aimed at preventing a fast sender from overwhelming a slow receiver.

Flow control can be helpful at reducing congestion, but it can't really solve the congestion problem. For example, suppose we connect a fast sender and fast receiver (eg., two Crays) using a 9.6 kbps line:

1. If the two machines use a sliding window protocol, and the window is large, the link will become congested in a hurry.

2. If the window size is small (e.g., 2 packets), the link won't become congested.

Note how the window size limits the total number of packets that can be in transmission at one time.

Flow control can take place at many levels:

· User process to user process (end-to-end).

· Host to host. For example, if multiple application connections share a single virtual circuit between two hosts.

* Router to router. For example, in virtual circuits.


Peter Cole 2001

Flow SpecificationsTraffic shaping is most effective when the sender, receiver, and subnet all agree to it.

To get agreement, it is necessary to specify the traffic pattern in a precise way using a flow specification. A flow specification consists of a data structure that describes both the pattern of the injected traffic and the quality of service desired by the applications.

A flow specification can apply either to the packets sent on a virtual circuit, or to a sequence of datagrams sent between a source and a destination (or even to multiple destinations).


Peter Cole 2001

Congestion Control in Virtual Circuit Subnets

· One technique that is widely used to keep congestion that has already started from getting worse is admission control.

· The idea is simple: Once congestion has been signalled, no more virtual circuits are set up until the problem has gone away.

· Thus, attempts to set up new transport layer connections fail. Letting more people in just makes matters worse.

· An alternative approach is to allow new virtual circuits but carefully route all new virtual circuits around problem areas.


Peter Cole 2001

· virtual circuits can also use a form of flow specification

· the subnet will typically reserve resources along the path when the circuit is set up.

· These resources can include table and buffer space in the routers and bandwidth on the lines.

· This kind of reservation can be done all the time as standard operating procedure, or only when the subnet is congested.

· A disadvantage of doing it all the time is that it tends to waste resources.

· The price of the congestion control is unused bandwidth.


Peter Cole 2001

Traffic Shaping

Used in ATM subnets

Used to control burstiness of traffic

Much better to have a constant flow of data than feast or famine situation

Leaky Bucket

based on the principal of a bucket filled with water with a constant drip representing the flow - continuous and regular

Implementation requires the sender to have a internal out queue

One cell can be transmitted per clock tick and consequently it quietens the network

if the algorithm is used in a non-ATM network, ie. Variable length packets then restrict the number of bytes being transmitted rather than packets or any unit of data


Peter Cole 2001


Peter Cole 2001

Token Bucket Algorithm· The leaky bucket algorithm enforces a rigid output pattern at

the average rate.

· The Token Bucket algorithm allows the output to speed up when bursty traffic enters the bucket

· Basically, tokens are placed in the leaky bucket at a steady rate.

· A packet must first take a token and destroy it before it can be transmitted.

·


Peter Cole 2001

The leaky bucket algorithm does not allow idle hosts to save up permission to send large bursts later.

The token bucket algorithm does allow saving, up to the maximum size of the bucket, n. This property means that bursts of up to n packets can be sent at once, allowing some burstiness in the output stream and giving faster response to sudden bursts of input.

· Token bucket algorithm throws away tokens when the bucket fills up but never discards packets.

· Leaky bucket algorithm discards packets when the bucket fills up.


Peter Cole 2001

The leaky bucket and token bucket algorithms can also be used to smooth traffic between routers

A token bucket regulating a host can make the host stop sending when the rules say it must.


Peter Cole 2001

Telling a router to stop sending while its input keeps pouring in may result in lost data.

The implementation of the basic token bucket algorithm is just a variable that counts tokens. The counter is incremented by one every AT and decremented by one whenever a packet is sent. When the counter hits zero, no packets may be sent. In the byte-count variant, the counter is increment by k bytes every AT and decremented by the length of each packet sent.

A potential problem with the token bucket algorithm is that it allows large bursts even though the maximum burst interval can be regulated

One way to get smoother traffic is to put a leaky bucket after the token bucket. The rate of the leaky bucket should be higher than the token bucket's p but lower than the maximum rate of the network.


Peter Cole 2001

Load Shedding

When a router becomes overwhelmed by incoming packets it can use dump packets. This is load shedding and is another method of congestion control.

The router could dump the packets at random but most likely it will check the application that the packet belongs to and make a decision based on that packet.

Consider that dumping packets that are part of a connection oriented application, such as ftp, could cause possible re-transmission of packets already in the buffer of the router. Older packets are more valuable here so dumping newer ones is likely to cause less re-transmission.

For a video application dumping older packets is more acceptable than dumping newer ones.

Dumping newer ones in fovour of older ones can cause visual oddities (eg people moving backwards in a scene)

Another problem with video or sound transmission is Jitter.

Constant transmission rates are much more preferable than variable rates.


Peter Cole 2001

Negotiation for a average transmission rate can be made.

If a router receives a packet that is in front of its schedule it will delay that packet until it can transmit it on schedule

If a router receives a packet that is behind its schedule it will transmit that packet as soon as possible to attempt to maintain the agreed transmission rate.


Peter Cole 2001

Choke Packets· Choke packets can be used in both virtual circuit and

datagram subnets.

· Routers can monitor the level of congestion around them, and when congestion is present, they can send choke packets to the sender that say "slow down".

· How can a router measure congestion? · A router might estimate the level of congestion by

measuring the percentage of buffers in use, line utilisation, or average queue lengths.

· Advantage: · Dynamic. Host sends as much data as it wants, the

network informs it when it is sending too much.

· Disadvantages:1. Difficult to tune. By how much should a host slow down?

depends on how much traffic the host is sending, how much of the congestion it is responsible for, and the total capacity of the congested region. Such information is not readily available in practice.

2. After receiving a choke packet, the sending host should ignore additional choke packets for a short while because packets currently in transmission may generate additional choke packets. How long? Depends on such dynamic network conditions as delay.


Peter Cole 2001

Spanning Tree BridgesSpanning tree bridges were designed with transparency as a primary goal. A customer should be able to buy a bridge, insert it between two networks, and have everything work correctly with no hardware, software, or configuration changes on either hosts or existing bridges. How do they work?

1. Each bridge maintains a table that maps destination addresses to the outgoing interface. (Analogous to routing tables in routers.)

2. Bridge operates in promiscuous mode, reading every frame on each of its connected LANS, and the routing decision is made as follows:

(a) Extract the source and destination address from the frame, and find the corresponding table entries for each address.

(b) If the two table entries point to the same interface, discard the frame. Why? If both pointers point to the same interface, both the sender and recipient are on the same local network (as far as we can tell), and the frame doesn't need to be forwarded.


Peter Cole 2001

(c) Otherwise, if the two pointers are different, send the frame out on the LAN given by the routing entry for the destination address.

(d) If the destination is not in the table, flood the frame on all interfaces (except the one on which it arrived). We don't know where the destination is, so let's be conservative and send it everywhere. That way we can be sure that the packet traverses the LAN on which the destination resides.

3. Bridges use backward learning to build tables. They determine which LAN to use to reach destination X by recording the interface on which frames having source address X arrive on.

4. The table is a cache; un-referenced entries are periodically flushed, allowing machines to be moved from one LAN to another.

The above approach works only for tree-structured topologies. Why? Frames will loop forever (there is no time-to-live field in LAN frames) if there are multiple distinct paths between any two bridges. To handle arbitrary topologies, bridges use a special protocol to build a spanning tree:


Peter Cole 2001

1. Bridges that are not part of the spanning tree are unused. That is, they are specifically excluded from the tree and do not forward packets. They are available for backup, however, should one of the other bridges or LANs fail.

2. Bridges periodically rebuild tables. They regularly exchange topology information, allowing them to detect the failure of a bridge or LAN. When a bridge or link that is part of the spanning fails, a new spanning tree is constructed.

Advantages:Easy, to use. Just install the bridges. No software changes are needed hosts.

Disadvantages:

1. Does not support multipath routing. By definition, only the bridges that belong to the spanning tree are used.

2. The path between any two hosts may not be the optimal path. An optimal path may traverse a bridge that is not part of the spanning tree and cannot be used.

3. Broadcast and multicast frames must be flooded in all cases.


Peter Cole 2001

Source Routing BridgesSource routing bridges take a completely opposite approach from spanning tree bridges..

1. They are not transparent. Hosts treat frames sent locally differently from those sent through bridges. Conceptually, the sending host specifies a road map saying which bridges the frame must go through to reach its destination.

2. Each LAN is assigned a 16-bit LAN number, and each bridge on a LAN is assigned a 4-bit bridge number. The numbers must be unique and are set by the network administrator

3. Each frame carries a source route listing the path the frame is to take. The path consists of a sequence of [LAN number, bridge number] pairs.

4. Sending hosts (rather than bridges) responsible chooses the source route. Host selects paths by broadcasting (flooding) special discovery frames. A discovery frame includes space for each bridge to add its number to the recorded path.

5. Eventually, a discovery frame reaches the destination host , which returns it to the sender. Each returned message contains a viable path, and the sending host chooses the shortest one.


Peter Cole 2001

6. How many frames are generated? Unfortunately, the discovery process leads to frame explosion. The destination may receive an exponential number of copies of the original frame.

Advantages:

uses the optimal route. Also can make use of multiple paths to same destination.

Because paths aren't required to always lie along the spanning tree, better use of resources.

Disadvantages:

1. Not transparent to hosts; hosts must participate in source routing. This is a significant disadvantage.

2. Installing new bridges non-trivial. Specifically, a system administrator must assign LAN numbers and bridge numbers. Improper configuration leads to disaster.

3. Each host must detect bridge failure on its own (eg., using time-outs). With spanning tree bridges, the bridges hold


Peter Cole 2001

that responsibility, and once they have reconfigured, all hosts start using the new path at the same time.

Not surprisingly, IBM supports source routing bridges, while DEC supports spanning tree bridges. IBM markets token ring networks, while DEC has always been big on Ethernets.


Peter Cole 2001

Multi Protocol RoutersMulti Protocol Routers (sometimes referred to as gateways) are packet switches that operate at the network layer (level 3). They are quite often used to connect organisations, ie the lines coming in to them are often owned by different entities. These “Gateways” often are found at political boundaries and on company LAN boundaries

Operating at the network level gives Multi Protocol Routers increased flexibility compared to bridges in terms of

1. Translating addresses between dissimilar networks.

2. Fragmenting large packets for transmission across networks that carry only small maximum packet lengths.

3. Selecting an appropriate path through the subnet.

4. Enforcing policies (eg., don't forward any local packets off of this network).

Because Multi Protocol Routers do more things than bridges, they generally run slower than bridges. One issue that arises with Multi Protocol Routers is who owns them. Typically, bridges connect LANs of one organisation, and the issue does not arise there. The ownership question is important because someone has to be responsible for the Multi Protocol Routers operation and dual ownership frequently leads to finger pointing when something goes wrong.


Peter Cole 2001

1. The sending host opens a virtual circuit, but a circuit goes through gateway hops rather than router hops.

2. Any two neighbouring Multi Protocol Routers at the internetworking level must be connected to a common network.

3. Regular router-based virtual circuits connect neighbouring Multi Protocol Routers on the same physical network).

4. The end-to-end virtual circuit is a concatenation of individual virtual circuits through each of the networks along the path.

Connectionless internets operate just as connectionless networks. A host sends a packet to a neighbouring gateway, which forwards it the next gateway, and so forth. Just as with connectionless networks, Multi Protocol Routers make only a best-effort attempt at delivering the packet.

When a gateway receives a packet, it selects the interface to send the packet out on and encapsulates the packet using the local data link layer format. As a packet moves from gateway to gateway, it is repeatedly encapsulated and un-encapsulated as it travels across each network.


Peter Cole 2001

IP ProtocolThe goal of IP is to interconnect networks of diverse technologies and create a single, virtual network to which all hosts connect.

Hosts communicate with other hosts by handing datagrams to the IP layer; the sender does not worry about the details of how the networks are actually interconnected.

IP provides unreliable, connectionless delivery service.

IP defines a universal packet called an Internet Datagram

Datagrams contain the following fields:Version number (4-bits):


Peter Cole 2001

· Including a version number allows a future version of IP be used along side the current version, facilitating migration to new protocols.

Header length (4-bits): · Length of the datagram header (excluding data) in 32-bit

words. The minimum value is 5, maximum header length is 60 bytes. In practice, the length field is used to locate the start of the data portion of the datagram.

Type of Service:· basically what service do you require - fast, error free

etc. using combination of D,T and R bits (delay, throughput and reliability). Commonly not used today.

Total length (16-bits): · Total length of the IP datagram (in bytes), including

data and header. Maximum datgram size 65,535 bytes. The size of the data portion of the datagram is the total length minus the size of the header.

Identification:· ID of the datagram. All fragments of the same datagram

have the same ID no.

Fragment offset (13-bits), Flags (3-bits), Identifier (16-bits):

· These three fields are used for fragmentation and reassemble. DF don’t fragment - MF more fragments to come .

· Gateways are free to fragment datagrams as needed, and hosts are required to reassemble fragments before passing complete datagrams to the higher layer


Peter Cole 2001

protocols. Each fragment contains a complete copy of the original datagram's header plus some of the data .

Time-to-live (8-bits):· A hopcount that is decremented by each gateway.

Should the hopcount reach 0, discard the datagram. Originally, the time-to-live field was intended to reflect real time. In practice, it is now a hopcount. The time-to-live field squashes looping packets.

Protocol (8-bits): · What type of data the IP datagram carries (e.g., TCP,

UDP, etc.).

Header Checksum (16-bits): · A checksum of the IP header (excluding data).

Source address (.32-bits):· Original sender's address.

Destination address (32-bits):· Datagram's ultimate destination..


Peter Cole 2001

Options:


Peter Cole 2001

Network Byte Order

One problem that arises when interconnecting different machines is that different machines represent integers in different ways:

· Big Endian machines such as SPARC computers store the most significant byte of a 32-bit integer in the lowest memory address of the word (e.g. to the left).

· That is the transmission is left to right

· Little Endian machines such as Pentiums store the most significant byte at the highest address.

· That is the transmission is right to left

The Internet defines a network standard byte order that is used when referring to the fields of Internet datagrams, and the Internet specifies the use of Big Endian form.

Therefore a pentiums header needs to be converted before it goes out on the wire and visa versa when a header comes of the wire.


Peter Cole 2001

Internet AddressingHost identifiers are classified as names, addresses, or routes, where: A name suggests what objectwe want. An address specifies where the object is. A route tells us how to get to the object.

In the Internet, names consist of human-readable strings such as eve, percival, or gwen.cs.purdue.edu. Addresses consist of compact, 32-bit identifiers. Internet software translates names into addresses; lower protocol layers ajways uses addresses rather than names. Internet addresses are hierarchical, consisting of two parts:

network: The network part of an address identifies which network a host is on. Conceptually,each LAN has its own unique IP network number.

local: The local part of an address identifies which host on that network.

Later, we'll examine a technique called subnetting that adds a third level to the hierarchy. With subnetting, the local part


Peter Cole 2001

may consist of a "site" (e.g., SUNY-Albany, or GE), which is further broken down in to local network number, local host.

Conceptually, the Internet consists of a collection of physical networks, each of which is assigned a unique number As datagrams travel from one gateway to another, each gateway routes the datagram based on the network number in the datagram's destination address.

Only the gateway on the same network as the destination uses the local part of the address in forwarding a datagram.

That is, when the datagram reaches a gateway that connects to the destination address, the gateway uses the local part of the address to forward the datagram to the appropriate host.


Peter Cole 2001

The Internet designers were unsure whether the world would evolve into a few networks withmany hosts (e.g., large networks), or many networks each supporting only a few hosts (e.g., small networks).

· Internet addresses handle both large and small networks. Internet address are four bytes in size, where:

· Class A addresses start with a "O" in the most significant bit, followed by a 7-bit network address and a 24-bit local part.

· Class B addresses start with a "10" in the two most significant bits, followed by a 14-bit network number and a 16-bit local part.

· Class C addresses start with a "110" in the three most significant bits, followed by a 22-bit network number and an 8-bit local part.

· Class D addresses start with a "1110" in the four most significant bits, followed by a 28-bit group number.


Peter Cole 2001

Internet addresses can also refer to broadcast addresses. The all l's address is used to mean "broadcast on this network". Of course, if the underlying network technology doesn't support broadcasting, one can't broadcast Internet datagrams either.

Network addresses are written using dotted decimal notation. Each address consists of 4 bytes, and each byte is written in decimal form.

· 134.115.64.51 (class B network)

· sunya: 128.204 (a network address)

· Broadcast: 255.255.255.255

· Internet addresses refer to network connections rather than hosts.

· Gateways, for example, have two or more network connections and each interface has its own IP address.

· NOTE - there is not a one-to-one mapping between host names and IP addresses.

· Internet addresses are hierarchical addresses. Datagrams are initially routed only by network number, and only the


Peter Cole 2001

gateway connected to the destination network uses the local part while performing the routing operation.


Peter Cole 2001

DNS - Domain Name System

· Lower-level protocol layers use compact 32-bit Internet addresses. People and programs rarely refer to hosts, mailboxes, and other resources by their binary network addresses.

· they normally use ASCII strings, such as [email protected].

· The network itself only understands binary addresses, so some mechanism is required to convert the ASCII strings to network addresses.

· In the original ARPANET, there was simply a file, hosts.txt, that listed all the hosts and their IP addresses


Peter Cole 2001

· To add a name or change an address required contacting the central administrator, updating the table, and distributing it to all the other sites.

· This solution worked at first because most sites had only a few machines, and the table didn't require frequent changes.

· The centrally-maintained table suffered from several drawbacks:

1. No two machines could use the same machine name.

2. As the Internet grew, changes to the database took days to weeks to take effect.

3. The central site (nic.ddn.mil, previously known as sri-nic.arpa) became congested with the increase in the number of sites retrieving copies of the current table.

4. The Internet grew at an astonishing rate

· The situation became totally unmanageable

· The Domain Name System (DNS) is a hierarchical, distributed naming system designed to cope with the problem of explosive growth:


Peter Cole 2001

1. It is hierarchical because the name space is partitioned into subdomains.

2. It is distributed because management of the name space is delegated to local sites. Local sites have complete control (and responsibility) for their part of the name space. DNS queries are handled by servers called name servers.

3. It does more than just map machine names to internet addresses. For example, it allows a site to associate multiple machines with a single, site-wide mailbox name.

· In the DNS, the name space is structured as a tree, with domain names referring to nodes in the tree.


Peter Cole 2001

· The tree has a root, and a fully-qualified domain name is identified by the components of the path from the domain name to the root.

· The DNS links data objects called resource records (RRs) to domain names.

· RRs contain information such as internet addresses or pointers to name servers.

· A resource record is a five tuple of

· The Domain-name tells the domain to which this record applies. Normally, many records exist for each domain and each copy of the database holds information about multiple domains.

· This field is thus the primary search key used to satisfy queries. The order of the records in the database is not significant. When a query is made about a domain, all the matching records of the class requested are returned.

· The Time-to-live field gives an indication of how stable the record is. Information that is highly stable is assigned a large value, such as 86400 (the number of seconds in 1 day). Information that is highly volatile is assigned a small value, such as 60 (1 minute).


Peter Cole 2001

· Note: the TTL here serves a completely different purpose than the one found in the IP header.

· The Type field tells what kind of record this is.


Peter Cole 2001


Peter Cole 2001

Name Servers

Name servers are the programs that actually manage the name space.

The name space is divided into zones of authority, and a name server is said to be authoritative for all domain names within its zone.

Name servers can delegate responsibility for a subdomain to another name server, allowing a large name space to be divided into several smaller ones.


Peter Cole 2001

Name servers are linked by pointers.

When a name server delegates authority for a subdomain, it maintains pointers to the name servers that manage the subdomain.

The DNS can resolve fully-qualified names by starting at the root and following pointers until reaching an authoritative name server for the name being looked up.

Note: The shape of the name space and the delegation of subdomains does not depend on the underlying topology of the Internet.


Peter Cole 2001

DNS QueriesWhen a client (application) has a name to translate, it sends a DNS query to a name server.

DNS queries (and responses) are carried within UDP datagrams.

There are two types of queries:

Recursive: The server resolves the name completely, even if it has to send additional queries to other servers in order to obtain the desired answer.

Iterative: If the name server can't answer the query, have it return a pointer to another name server that has more information. The client then sends the query to the other name server.


Peter Cole 2001

ARP: Address resolution Protocol

· Suppose we have two machines A and B connected to the same network, and A wants to send an internet datagram to B.

· A must know B's data link layer address in order to send frames to B.

· The problem of mapping Internet addresses to physical addresses is known as the address resolution problem.

Address resolution is complex for networks such as Ethernets:


Peter Cole 2001

· Each Ethernet device has its own unique address. Replacing a host's Ethernet card changes its physical address.

· Physical address are 6 bytes long, (48 bits) too large to encode within an Internet address (32 bits)

· new machines can be added to the network with no disruption of service. Thus, adding new hosts should not require reconfiguring existing hosts to inform them of the new machine.

The Address Resolution Protocol (ARP) is a protocol that allows hosts to dynamically map Internet addresses to physical addresses:

1. The requesting machine only needs to know the target machine's IP address.

2. It sends out a special ARP request frame using the Ethernet's broadcast capability. Thus, every machine on the LAN will receive the ARP request.

3. The ARP request asks "what is the Ethernet address of Internet address X"?

4. Each machine receives a copy of the broadcast message, and the machine having the desired IP address responds with its Ethernet address.


Peter Cole 2001

· a machine doesn't send out an ARP packet each time it wishes to send an IP datagram.

· each machine maintains a cache of recently used mapping’s, and an ARP request is only sent if the desired mapping is not already in the cache.

· ARP request packets also contain the sender's IP and Ethernet address pair.

· Why? To eliminate the need for a second ARP request.

· If machine A wishes to communicate with machine B, there is high probability that B will need A's Ethernet address as well.

· Upon receipt of an ARP request from a machine whose IP address is already in the local ARP cache, updates the information for that entry. This handles the case of a machine whose Ethernet address changes; ARP entries with the old value will be overwritten with the new value.

From a layering point of view, ARP sits below IP, but above the data link layer, as shown in Figure 7.

Conceptually, ARP consists of two parts: the software responsible for finding the physicaj address of an IP address (e.g., a client), and the software responsible for answering ARP requests from other machines (e.g., a server)


Peter Cole 2001

Reverse ARP

· ARP handles the case of determining the hardware address that corresponds to an IP address.

· When is it necessary to map hardware addresses back into IP addresses?

· When a diskless machine first boots, it doesn't know its own IP address (and can't read it from a local disk!).

· How can a booting station get started?

· Have the booting client contact a server or any other machine to obtain its Internet address.

· The protocol that maps hardware addresses to Internet addresses is called Reverse ARP

· A RARP server maintains a database of physical address to Internet address mapping.

· Each network requires a RARP server.

one of the primary benefits of broadcasting: locating servers. However, because broadcasting is resource intensive and it should be used sparingly.


Peter Cole 2001

lecture 1 week 1 - murdoch universityftp.it.murdoch.edu.au/units/ict227/b527/lectures/b227… ·...

Documents