COP 5615: Distributed Operating System Principles | Weekly Assignment Week 2: 26-30 August
Architecture
There are multiple architectural styles when It comes to designing a system.
Layered Architecture: It consists of multiple layers where each layer provides a different service. It
offers very good separation of concern for developing scalable, production grade, distributed, relatively
risk-free applications. Ex. Communication Protocols (TCP/IP has 4 layers).
Resource based architecture: System that follow the RESTful design criteria. The system is a collection
of resources which may be added, removed, retrieved and modified by applications via operations like
GET, POST, PUT, DELETE. Ex. Amazon AWS S3 (http://BucketName.s3.amazonaws.com/ObjectName).
Centralized system Architecture: It uses client/server architecture where one or more client nodes are
connected to a central server (can be on different machine). Clients follow request/reply model for using
services. Disadvantages are Its difficult to scale vertically and there is a single point of failure.
Decentralized Architecture: No single entity is responsible for receiving and responding to requests. No
single point of failure. Example Torrent network, blockchain and its applications like bitcoin. No single
entity/organization owns the bitcoin network. It can be-
Structured Peer to peer: Makes use of semantic-free Index (hash function) to assign values to every peer
and content where each peer is responsible for specific part of content. Ex chord (nodes in a ring).
Unstructured Peer to peer: Overlay links are created arbitrarily where each node maintains an ad hoc list
of neighbors. Is easy to add a new node. Searching for data is done by flooding (node passing request to
all neighbors) or Random walk (node passing request to randomly chosen node).
Hybrid Architecture: To provide tangible increase in data access speed an edge server system can be
used where servers are placed at edge of the network. Ex. Netflix uses this to reduce latency and
improve performance while distributing their content to their users.
Processes and Threads
Processors: are physical hardware. It has the capability to execute a series of instructions. We build
virtual processor on top of physical ones.
Process: is software processor in whose context a series of instructions can be executed.
Thread: is a minimal software processor which executes in context of a process. It has all the data
needed for instruction execution. Multiple threads can run under same process.
We use multiple threads to exploit parallelism on a multiprocessor and avoid needless blocking when
performing I/O leveraging context-switching.
Context Switching is the process of saving the context of one thread/process and loading the context of
another thread/process.
Process switching involves switching the memory address space like memory address, mapping, page
tables, kernel resources (might involve flushing cache). Thread switching involves switching abstractions
unique to thread like program counter and register state which is much cheaper.
So, Process context switching is expensive than thread context switching and should be avoided.
-Saurabh K Prasad (UFID: 3417-8993)