Technical Interviews: Difference between revisions

From David's Wiki
Line 52: Line 52:


====Load Balancers====
====Load Balancers====
* Purpose: Handle many connections and route them to request nodes
* Algorithms: Random, round robin, custom criteria
* Software: [https://www.haproxy.org/ HAProxy]
* Usually placed at the front of a distributed system
** Load balancers can send requests to other load balancers
* Challenges: Managing user session data
** Solutions: Caches, cookies, user data, URL rewriting
* Cons: Makes problem diagnosis difficult


====Queues====
====Queues====


==System Design==
==System Design==

Revision as of 21:20, 19 December 2019

Resources for Technical Interviews


Algorithms

Computer Systems

Web Architecture

Four Key Principles

  • Availability
  • Performance
  • Reliability
  • Scalability
  • Manageability
  • Cost

Services

  • Separate each functionality into its own service so each can be scaled separately
(e.g. Reading and Writing can be two services which access the same file store)
  • Each service can then be managed separately.

Redundency

  • Replicate nodes for each service using a "shared-nothing architecture."
  • Each node should be able to operate independently.

Partitions

  • Also known as shards
  • Scaling vertically: Add more hard drives, memory
  • Scaling Horizontally: Add more nodes
  • Distribute data somehow: geographically, by type of user,...
  • Make sure you can identify server from id of data (e.g. adding it to hashing or maintaining an index)

Caches=

  • Place a cache on the request layer.
  • Cache in memory, not disk (e.g. memcached, redis)
Global Caches
  • If you have several request nodes, you can put a global cache between the request node and the database
Distributed Cache
  • Each request node holds a cache for a specific type of data
  • A request node may forward requests to other request nodes which have the corresponding cache
  • Cons: May be hard to remedy if a node goes down

Proxies

  • Used to filter, log, and modify requests
  • Collapsed Forwarding: collapse the same or similar requests and return the same result to clients
  • You can also collapse requests that are spatially close together on the database


Indexes

Load Balancers

  • Purpose: Handle many connections and route them to request nodes
  • Algorithms: Random, round robin, custom criteria
  • Software: HAProxy
  • Usually placed at the front of a distributed system
    • Load balancers can send requests to other load balancers
  • Challenges: Managing user session data
    • Solutions: Caches, cookies, user data, URL rewriting
  • Cons: Makes problem diagnosis difficult

Queues

System Design