Transcript Introduction - Suraj @ LUMS

CS 582 / CMPE 481
Distributed Systems
Communications (cont.)
Class Overview
•
•
•
•
•
•
What is a network?
Protocols
Layered architecture
Reference models
Middleware
IPC
–
–
–
–
Why IPC?
Design Considerations
Client-Server Communication
Group Communication
• RPC
Middleware
• Middleware is invented to provide common services and
protocols that can be used by many different applications:
– A rich set of communication protocols, but which allow different
applications to communicate
– Marshaling and unmarshaling of data, necessary for integrated
systems
– Naming protocols, so that different applications can easily share
resources
– Security protocols, to allow different applications to communicate
in a secure way
– Scaling mechanisms, such as support for replication and caching
• What remain are truly application-specific protocols
Interaction Model in Distributed Systems
• A principal goal of distributed systems
– a single system image => resource sharing
• concurrency
• Cooperation
– explicit communication is required between components
• Common form of interaction
– client-server interaction model
• interprocess communications (IPC)
Principal Functions
• Allows communication between components
(processes)
• Shields one process from failure of another
• Provides modularity by a well defined interface
mechanism
• hides distinction between local and remote
communications
Abstraction of Communication Subsystem
• Regards a network as another I/O device
– no abstraction beyond a transport protocol
• client and server are wholly responsible for message exchange over a given
transport mechanism (e.g. TCP/IP)
• primitives: write (send) and read (receive)
• Request and reply
– abstraction of message passing required to execute a procedure at a
server
• primitives: DoOperation, GetRequest, and SendReply
• Remote procedure call
– hides the separation between a client and a server
• makes invocation of a remote procedure at a server same as that of a local
procedure
Design Considerations in IPC
•
•
•
•
•
•
Data representation
Marshalling
Calling semantics
Addressing
Reliable delivery
Versatility
Data Representation
• Due to a heterogeneity property of distributed systems
• Things to consider
– data type representation
– byte ordering
– word boundary
• Policy
– conversion to the common data type agreed by both a sender and a
receiver
– a receiver makes it right (tagging)
– a sender negotiates with a receiver
• Parameter passing
• Examples: Sun XDR, ISO ASN.1, Xerox Courier
Marshalling
•
marshal·ling
1. To arrange or place (troops, for example) in line for a parade,
maneuver, or review.
2. To arrange, place, or set in methodical order.
3. To enlist and organize.
4. To guide ceremoniously; conduct or usher.
•
Marshalling
–
–
–
•
linearization of an operation invocation
convert the type of data into a common type (if needed)
pack linearized data into a message
Unmarshalling
–
–
reverse operation of marshalling at the receiving peer
upcalls
Calling Semantics
• Remote operation invocation may need a different
behavior than a local one
– sender may not need a result either at all or immediately
– sender may want to do other operations in parallel (e.g.
multithreading)
• Semantics
– synchronous: same as local operation invocation
(blocked)
• timeout is used to avoid indefinite wait
– delayed synchronous: the sender gets a reply later
– asynchronous: the sender does not need a reply
• Blocking vs. Non-blocking
– R/RR/RRA [Spector 1982]
Addressing
• Identification of communication peers
• Location-independent identifiers
– functional addressing: port in Amoeba
– globally unique identifier: UUID in NCA
• Types of destination identifiers
– Port
– mailbox
Reliable Delivery
• Reliable Delivery
• End-to-end argument [Saltzer, Reed, Clark]
– error recovery in lower levels of protocols is only useful for
purposes of increasing efficiency
• protocols for application-level end-to-end checking are always required
• Possible causes for retransmission (client -> server)
– server still working on it
– server crashes
– request gets lost
• IPC should provide some level of failure transparent but
also failure visibility (accurate report of failures)
Reliable Delivery (cont.)
• Fault management based on execution semantics
– at-least-once: idempotent
– at-most-once: session
– exactly-once
• Retransmission mechanisms
–
–
–
–
reply can be considered an acknowledgement
selective retransmission: sequence number
explicit acknowledgement
retransmission timer
Versatility
• Types of IPC
–
–
–
–
remote operation
bulk data transfer
group communication
continuous media
• Rate control
– stop-and-wait
– sliding window
– Blasting
• Multipacket
• QoS
– real-time: retry is better than retransmission
– stream-oriented: steady and low delay but packet lost is ok
Examples: Unix IPC
• Socket provides a message passing mechanism
–
–
–
–
–
send/sendto and recv/recvfrom primitives
select allows non-blocking semantics (with timeout)
addressing: host id + port number
scatter-gather mechanism for large datagram
QoS
• TCP: reliable transmission
• UDP: unreliable but fast transmission
Group Communication
• Purposes of multicast messages
–
–
–
–
fault tolerance based on replicated services
locating objects in distributed services
better performance through replicated data
multiple update
• Atomicity
– atomic multicast
• all or nothing
• synchronization among members (e.g. join or leave of members)
– Reliable
• best effort to deliver messages but no guarantee
• computation or query
Ordering in Group Communication
• Why ordering is concerned?
– concurrent execution of update requests at replicas may result in
inconsistency among replicated data
– serial equivalence of update requests is required
• expense of ordering should also be considered
• Ordering requirements
– total ordering
• requests are processed in the same order at all replicas
– causal ordering
• causally related requests are only ordered at all replicas
– sync ordering
• requests are ordered in sync before or after a certain request at all replicas
Ordering in Group Communication (cont.)
• Request handling at replicas
– every request is held-back until ordering constraints can
be met
– request is defined to be stable at a replica once no request
from a client and bearing a lower unique identifier can be
subsequently delivered to replica; that is, all prior
requests have been processed
• Request ordering implementation
– group communication
– exchanging gossip messages among replicas