Technical Interviews

Resources for Technical Interviews

Algorithms

Main article: Interview Algorithms

In general, consider the following approaches

• Brute-force and searching
• Dynamic Programming
• Greedy (sorting + take the best at each step)
• Linear searching (with 2 pointers)
  • Usually the answer if you're searching for a connected segment along an array
  • E.g. max contiguous subarray

Computer Systems

Web Architecture

• Scalable Web Architecture and Distributed Systems

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

• Indexes are used to quickly find small data in large datasets
• Indexes are stored in memory, data separated across several servers
• Possibly many layers of 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

• Schedule client tasks in a queue, acknowledge task received immediately
• Queues protect from service outages and failures
• Software: RabbitMQ, ActiveMQ

Misc

• Steve Yegge Get that job at Google