System Design Introduction For Interview

System Design Introduction For Interview.md

Introduction For Interview

YouTube - Video tutorial by Tushar Roy

Things to do

• Ask good questions
  • What features to work on
  • How much to scale
• Don't use buzzwords
• Clear and organized thinking
• Drive discussions (80-20 rule)

Things to consider

• Features
• Define APIs
• Availability
  • What if something goes down
• Latency Performace
  • Cache
• Scalability
• Durability
  • Data stored securely
  • Not lost
  • Not compromised
  • Matter in some interviews
• Class Diagram
  • OO Priciples
• Security & Privacy
• Cost effective

Concepts

• Vertical & Horizontal scaling
  • Vertical - Cap at how much for 1 host
  • Horizontal - Infinite add more hosts
    • Has to deal with distributed system challenges
    • Preferred over vertical
• CAP Theorem
  • Consistency - read has most recent write
  • Availability - get data back irregardless of most recent or not
  • Partition Tolerance - bw 2 nodes packets could be dropped
  • Only achieve 2/3
  • Traditional Consistency > Availability
  • NoSQL Availability > Consistency
• ACID - Atomic Consistent Isolated Durable
  • Relational
• BASE - Basic Availability Soft-State Eventual consistency
  • NoSQL
  • If use NoSQL know which part of ACID sacrifice
• Partitioning / Sharding Data
  • Store in different servers
  • Consistent Hashing
• Optimistic vs Pessimistic Locking
• Strong vs Eventual consistency
  • Strong - always see latest write
    • Relation more
  • Eventual - see some, eventually will see latest
    • NoSQL, can choose
    • Higher availability
• RelationalDB vs NoSQL
  • rDB - ACID
  • NoSQL - Scales better, better availability
• Types of NoSQL
  • Key Value
  • Wide Column
  • Document Based
    • Semi-structured
    • XML, JSON
  • Graph Based
• Caching
  1. Every node does own cache
  2. Distributed cache - sharing
  • Considerations
    • Cache cannot be source of truth
    • Small - in memory
    • Eviction policy? around cache
• Data center / Racks / hosts
  • Latency cross hosts, DC
• CPU/Memory/HD/Bandwidth
  • Limited resources
• Ramdom vs sequential read/write on disk
  • Design around sequential since faster
• HTTP vs HTTP2 vs websockets
  • HTTP2 improve
    • Multiple requests or single
  • websocket - fully bidirectional communication
• TCP/IP Model
• IPV4 vs IPV6
  • ipv4 = 32 bit addresses
  • ipv6 = 128 bit addresses
• TCP vs UDP
  • TCP - reliable
    • documents
  • UDP - unreliable
    • Faster
    • Video
• DNS Lookup
  • Translate address -> ip address
  • Caching
• HTTPS & TLS
• Public key infrastructure & certificate authority
• symmetric vs asymmetric encryption
  • asymmetric - private / public encryption
    • More computation so better for smaller data
  • symmetric - AES
• Load Balancer
  • Sit in front service and delegate to node behind server
  • Round Robin or load average
  • L4
    • Both client and destination & port
  • L7
    • HTTP URI for routing
• CDNs & Edge
  • Content Delivery Network
    • Data close to streaming point
    • Improve performance and latency
  • Edge
    • Processing close to end-user
• Bloom Filters & Count-min sketch
  • Bloom Filters
    • Determine if element is part of set
    • False positives but never false negatives
    • Space efficient
  • Count-min sketch
    • Count frequency of events
    • Close to answer w/ some error rate
• Paxos - Consensus over distributed hosts
  • Paxos solve: Leader election amonst hosts
• Design Patterns & OO
• Virtual Machines & Containers
  • containers - run application & dependencies in isolated env.
• Publishers-Subscribers over Queue
  • Publish -> Queue -> Subscriber
  • Customer facing should not expose to PubSub system
• MapReduce
  • Distributed mass parallel big data
  • Map - filtering and sort
  • Reduce - Summarizing data
• Multithreading, concurrency, locks, synchronization, CAS

Tools

• Cassandra
  • Wide-column high-scalable DB
  • Simple key value
  • Time series data
  • Rows w/ lots of colums
  • Eventual and strong consistency
  • Hashing to shard data
  • Gossiping to keep nodes informed
• MongoDB
  • JSON
  • ACID at Document
  • Scale
• MySQL
  • Traditional
  • ACID
  • Master Slave
• Memcached
  • Distributed cache
  • Hold data in memory
  • Key value
• Redis
  • Key value
  • Cluster
• Zookeeper
  • Centralized config management
  • Leader election
  • Distributed locking
  • Scale for reads but not writes
  • In memory
• Kafka
  • Fault tolerant , highly available queue
  • PubSub or stream
  • Deliver messages once
  • Keep ordered in partition
• NGINX & HA Proxy
  • Load balancers
  • Efficient
• Solr & Elastic Search
  • Full text search
• Blobstore like Amazon S3
  • Picture & file storage
• Docker
  • Containers
  • Kubernetes & Mesos
    • Manage & coordinate containers
• Hadoop/Spark
  • Hadoop - uese map reduce
  • Spark - faster Hadoop since in-memory
  • HDFS
    • Java based
    • Distributed & fault tolerant
  • Hadoop relies on HDFS for processing