Group:
information
Topic:
communication protocols
Topic:
dependency analysis
Topic:
event time
Topic:
implementing distributed systems and applications
Topic:
Internet
Topic:
model checker
Topic:
reliable broadcast
Topic:
reliability of distributed systems
Topic:
resourceful, redundant systems for reliability
Topic:
timestamps
Topic:
World-Wide Web
| |
Summary
Communication can not avoid error; it must handle errors reliabily and efficiently. Entropy is a measure of information. For reliable communication, the entropy of the source must be less than the channel capacity. The Internet's primary goal is survivability.
Reliability analysis may be simplified by an end-to-end approach. With sufficient redundancy, errors can be corrected. Weak properties such as at most once delivery may be more practical then strong properties such as causal, totally ordered communication. (cbb 8/06)
Subtopic: end-to-end error recovery, authorization 
| Quote: while end-to-end error recovery is sufficient, it requires a cheap test for success and may cause severe performance problems under heavy load [»lampBW10_1983]
| | Quote: the file transfer program must ensure end-to-end reliability; the communication channel is one link in a long chain [»saltJH11_1984]
| | Quote: fate-sharing--can loss state about an entity only if lose the entity itself; e.g., store synchronization information with the hosts and not the gateways, trust the host instead of the network
| | Quote: limited usefulness for causally and totally ordered communication support (CATOCS); such state problems violate the end-to-end argument [»cherDR12_1993]
| | Quote: end-to-end authorization across administrative, network, abstraction, and protocol boundaries; uses SPKI [»howeJ_2000]
|
Subtopic: entropy
| Quote: fundamental theorem for a discrete channel with noise; if source entropy less than channel capacity, average frequency of error is arbitrarily small [»shanCE7_1948]
| | Quote: if the entropy of a source is not more than a channel's capacity then have arbitrarily reliable transmission [»weinGM_1979]
| | Quote: define information by physical characteristics; the rate of transmission is limited by energy considerations [»hartRV7_1928]
| | Quote: rate of information transmission is proportional to the width of frequency-range times time delta
|
Subtopic: verification
| Quote: experimentally derived protocols for TCP/IP and Sockets identified many behavioural anomalies and explicit OS version dependencies [»bishS1_2006]
|
Subtopic: authorized channels
| Quote: EROS allows transmission of capabilities across authorized communication paths; this does not limit security [»shapJS1_2002]
|
Subtopic: internet reliability
| Quote: the ordering of goals for the Internet is important: survivability, multiple types of services, ..., accountability [»clarDD8_1988]
| | Quote: the Internet is a packet switched communications facility; a network of networks connected by gateways using store and forward packet forwarding
| | Quote: except for total partition, the Internet masks all transient failures; must protect the state about an on-going conversation [»clarDD8_1988]
| | Quote: the Internet separates TCP from IP to allow other transports that optimize delay or bandwidth instead of reliability [»clarDD8_1988]
| | Quote: TCP provides a reliable sequenced data stream while IP provides the basic building block, the datagram
|
Subtopic: ethernet
| Quote: an Ethernet interface board can have serious hardware problems yet perform well; due to high-level protocols for fault handling [»liskB10_1981]
|
Subtopic: error correction
| Quote: the tea-leaf reader CRC algorithm for error-detection processes a byte at a time with a look-ahead table; efficient [»grifG7_1987]
| | Quote: Weighted Sum Codes are as good as CRC and easier to implement; also better than other checksum codes [»mcauAJ2_1994]
|
Subtopic: queuing
| Quote: input messages stored in a queue, only removed when completely processed [»giffDK12_1985]
|
Subtopic: kernel IPC
| Quote: all client/server communication via V kernel IPC facilities; for protection and autonomy [»cherDR4_1984]
|
Subtopic: at most once
| Quote: despite lost messages and crashed servers, Amoeba guarantees that all messages are delivered at most once [»taneAS12_1990]
|
Subtopic: problems with causal, totally ordered communication
| Quote: atomic message delivery needed for causally and totally ordered communication (CATOCS) [»cherDR12_1993]
| | Quote: CATOCS message ordering does not enforce casual dependencies due to a shared database or external environment; e.g. a sensor [»cherDR12_1993]
| | Quote: CATOCS message ordering does not enforce ordering of groups of messages; e.g., a group update [»cherDR12_1993]
| | Quote: CATOCS message ordering does not enforce casual memory, linearizability, or serializability [»cherDR12_1993]
| | Quote: CATOCS message ordering delays unrelated messages; happens-before does not imply dependency
| | Quote: CATOCS message ordering is expensive; e.g., quadratic growth in message buffering as number of processes increase [»cherDR12_1993]
| | Quote: CATOCS message ordering easily replaced by straight-forward, state-based solutions [»cherDR12_1993]
|
Related Topics
Group: information (46 topics, 1160 quotes)
Topic: communication protocols (62 items)
Topic: dependency analysis (34 items)
Topic: event time (45 items)
Topic: implementing distributed systems and applications (41 items)
Topic: Internet (16 items)
Topic: model checker (49 items)
Topic: reliable broadcast (29 items)
Topic: reliability of distributed systems (35 items)
Topic: resourceful, redundant systems for reliability (38 items)
Topic: timestamps (19 items)
Topic: World-Wide Web
(42 items)
|
