Home

Awesome

English日本語简体中文繁體中文 | العَرَبِيَّة‎বাংলাPortuguês do BrasilDeutschελληνικάעבריתItaliano韓國語فارسیPolskiрусский языкEspañolภาษาไทยTürkçetiếng ViệtFrançais | Add Translation

The System Design Primer

<p align="center"> <img src="http://i.imgur.com/jj3A5N8.png"/> <br/> </p>

Motivation

Learn how to design large-scale systems.

Prep for the system design interview.

Learn how to design large-scale systems

Learning how to design scalable systems will help you become a better engineer.

System design is a broad topic. There is a vast amount of resources scattered throughout the web on system design principles.

This repo is an organized collection of resources to help you learn how to build systems at scale.

Learn from the open source community

This is a continually updated, open source project.

Contributions are welcome!

Prep for the system design interview

In addition to coding interviews, system design is a required component of the technical interview process at many tech companies.

Practice common system design interview questions and compare your results with sample solutions: discussions, code, and diagrams.

Additional topics for interview prep:

Anki flashcards

<p align="center"> <img src="http://i.imgur.com/zdCAkB3.png"/> <br/> </p>

The provided Anki flashcard decks use spaced repetition to help you retain key system design concepts.

Great for use while on-the-go.

Coding Resource: Interactive Coding Challenges

Looking for resources to help you prep for the Coding Interview?

<p align="center"> <img src="http://i.imgur.com/b4YtAEN.png"/> <br/> </p>

Check out the sister repo Interactive Coding Challenges, which contains an additional Anki deck:

Contributing

Learn from the community.

Feel free to submit pull requests to help:

Content that needs some polishing is placed under development.

Review the Contributing Guidelines.

Index of system design topics

Summaries of various system design topics, including pros and cons. Everything is a trade-off.

Each section contains links to more in-depth resources.

<p align="center"> <img src="http://i.imgur.com/jrUBAF7.png"/> <br/> </p>

Study guide

Suggested topics to review based on your interview timeline (short, medium, long).

Imgur

Q: For interviews, do I need to know everything here?

A: No, you don't need to know everything here to prepare for the interview.

What you are asked in an interview depends on variables such as:

More experienced candidates are generally expected to know more about system design. Architects or team leads might be expected to know more than individual contributors. Top tech companies are likely to have one or more design interview rounds.

Start broad and go deeper in a few areas. It helps to know a little about various key system design topics. Adjust the following guide based on your timeline, experience, what positions you are interviewing for, and which companies you are interviewing with.

ShortMediumLong
Read through the System design topics to get a broad understanding of how systems work:+1::+1::+1:
Read through a few articles in the Company engineering blogs for the companies you are interviewing with:+1::+1::+1:
Read through a few Real world architectures:+1::+1::+1:
Review How to approach a system design interview question:+1::+1::+1:
Work through System design interview questions with solutionsSomeManyMost
Work through Object-oriented design interview questions with solutionsSomeManyMost
Review Additional system design interview questionsSomeManyMost

How to approach a system design interview question

How to tackle a system design interview question.

The system design interview is an open-ended conversation. You are expected to lead it.

You can use the following steps to guide the discussion. To help solidify this process, work through the System design interview questions with solutions section using the following steps.

Step 1: Outline use cases, constraints, and assumptions

Gather requirements and scope the problem. Ask questions to clarify use cases and constraints. Discuss assumptions.

Step 2: Create a high level design

Outline a high level design with all important components.

Step 3: Design core components

Dive into details for each core component. For example, if you were asked to design a url shortening service, discuss:

Step 4: Scale the design

Identify and address bottlenecks, given the constraints. For example, do you need the following to address scalability issues?

Discuss potential solutions and trade-offs. Everything is a trade-off. Address bottlenecks using principles of scalable system design.

Back-of-the-envelope calculations

You might be asked to do some estimates by hand. Refer to the Appendix for the following resources:

Source(s) and further reading

Check out the following links to get a better idea of what to expect:

System design interview questions with solutions

Common system design interview questions with sample discussions, code, and diagrams.

Solutions linked to content in the solutions/ folder.

Question
Design Pastebin.com (or Bit.ly)Solution
Design the Twitter timeline and search (or Facebook feed and search)Solution
Design a web crawlerSolution
Design Mint.comSolution
Design the data structures for a social networkSolution
Design a key-value store for a search engineSolution
Design Amazon's sales ranking by category featureSolution
Design a system that scales to millions of users on AWSSolution
Add a system design questionContribute

Design Pastebin.com (or Bit.ly)

View exercise and solution

Imgur

Design the Twitter timeline and search (or Facebook feed and search)

View exercise and solution

Imgur

Design a web crawler

View exercise and solution

Imgur

Design Mint.com

View exercise and solution

Imgur

Design the data structures for a social network

View exercise and solution

Imgur

Design a key-value store for a search engine

View exercise and solution

Imgur

Design Amazon's sales ranking by category feature

View exercise and solution

Imgur

Design a system that scales to millions of users on AWS

View exercise and solution

Imgur

Object-oriented design interview questions with solutions

Common object-oriented design interview questions with sample discussions, code, and diagrams.

Solutions linked to content in the solutions/ folder.

Note: This section is under development

Question
Design a hash mapSolution
Design a least recently used cacheSolution
Design a call centerSolution
Design a deck of cardsSolution
Design a parking lotSolution
Design a chat serverSolution
Design a circular arrayContribute
Add an object-oriented design questionContribute

System design topics: start here

New to system design?

First, you'll need a basic understanding of common principles, learning about what they are, how they are used, and their pros and cons.

Step 1: Review the scalability video lecture

Scalability Lecture at Harvard

Step 2: Review the scalability article

Scalability

Next steps

Next, we'll look at high-level trade-offs:

Keep in mind that everything is a trade-off.

Then we'll dive into more specific topics such as DNS, CDNs, and load balancers.

Performance vs scalability

A service is scalable if it results in increased performance in a manner proportional to resources added. Generally, increasing performance means serving more units of work, but it can also be to handle larger units of work, such as when datasets grow.<sup><a href=http://www.allthingsdistributed.com/2006/03/a_word_on_scalability.html>1</a></sup>

Another way to look at performance vs scalability:

Source(s) and further reading

Latency vs throughput

Latency is the time to perform some action or to produce some result.

Throughput is the number of such actions or results per unit of time.

Generally, you should aim for maximal throughput with acceptable latency.

Source(s) and further reading

Availability vs consistency

CAP theorem

<p align="center"> <img src="http://i.imgur.com/bgLMI2u.png"/> <br/> <i><a href=http://robertgreiner.com/2014/08/cap-theorem-revisited>Source: CAP theorem revisited</a></i> </p>

In a distributed computer system, you can only support two of the following guarantees:

Networks aren't reliable, so you'll need to support partition tolerance. You'll need to make a software tradeoff between consistency and availability.

CP - consistency and partition tolerance

Waiting for a response from the partitioned node might result in a timeout error. CP is a good choice if your business needs require atomic reads and writes.

AP - availability and partition tolerance

Responses return the most recent version of the data available on a node, which might not be the latest. Writes might take some time to propagate when the partition is resolved.

AP is a good choice if the business needs allow for eventual consistency or when the system needs to continue working despite external errors.

Source(s) and further reading

Consistency patterns

With multiple copies of the same data, we are faced with options on how to synchronize them so clients have a consistent view of the data. Recall the definition of consistency from the CAP theorem - Every read receives the most recent write or an error.

Weak consistency

After a write, reads may or may not see it. A best effort approach is taken.

This approach is seen in systems such as memcached. Weak consistency works well in real time use cases such as VoIP, video chat, and realtime multiplayer games. For example, if you are on a phone call and lose reception for a few seconds, when you regain connection you do not hear what was spoken during connection loss.

Eventual consistency

After a write, reads will eventually see it (typically within milliseconds). Data is replicated asynchronously.

This approach is seen in systems such as DNS and email. Eventual consistency works well in highly available systems.

Strong consistency

After a write, reads will see it. Data is replicated synchronously.

This approach is seen in file systems and RDBMSes. Strong consistency works well in systems that need transactions.

Source(s) and further reading

Availability patterns

There are two main patterns to support high availability: fail-over and replication.

Fail-over

Active-passive

With active-passive fail-over, heartbeats are sent between the active and the passive server on standby. If the heartbeat is interrupted, the passive server takes over the active's IP address and resumes service.

The length of downtime is determined by whether the passive server is already running in 'hot' standby or whether it needs to start up from 'cold' standby. Only the active server handles traffic.

Active-passive failover can also be referred to as master-slave failover.

Active-active

In active-active, both servers are managing traffic, spreading the load between them.

If the servers are public-facing, the DNS would need to know about the public IPs of both servers. If the servers are internal-facing, application logic would need to know about both servers.

Active-active failover can also be referred to as master-master failover.

Disadvantage(s): failover

Replication

Master-slave and master-master

This topic is further discussed in the Database section:

Availability in numbers

Availability is often quantified by uptime (or downtime) as a percentage of time the service is available. Availability is generally measured in number of 9s--a service with 99.99% availability is described as having four 9s.

99.9% availability - three 9s

DurationAcceptable downtime
Downtime per year8h 45min 57s
Downtime per month43m 49.7s
Downtime per week10m 4.8s
Downtime per day1m 26.4s

99.99% availability - four 9s

DurationAcceptable downtime
Downtime per year52min 35.7s
Downtime per month4m 23s
Downtime per week1m 5s
Downtime per day8.6s

Availability in parallel vs in sequence

If a service consists of multiple components prone to failure, the service's overall availability depends on whether the components are in sequence or in parallel.

In sequence

Overall availability decreases when two components with availability < 100% are in sequence:

Availability (Total) = Availability (Foo) * Availability (Bar)

If both Foo and Bar each had 99.9% availability, their total availability in sequence would be 99.8%.

In parallel

Overall availability increases when two components with availability < 100% are in parallel:

Availability (Total) = 1 - (1 - Availability (Foo)) * (1 - Availability (Bar))

If both Foo and Bar each had 99.9% availability, their total availability in parallel would be 99.9999%.

Domain name system

<p align="center"> <img src="http://i.imgur.com/IOyLj4i.jpg"/> <br/> <i><a href=http://www.slideshare.net/srikrupa5/dns-security-presentation-issa>Source: DNS security presentation</a></i> </p>

A Domain Name System (DNS) translates a domain name such as www.example.com to an IP address.

DNS is hierarchical, with a few authoritative servers at the top level. Your router or ISP provides information about which DNS server(s) to contact when doing a lookup. Lower level DNS servers cache mappings, which could become stale due to DNS propagation delays. DNS results can also be cached by your browser or OS for a certain period of time, determined by the time to live (TTL).

Services such as CloudFlare and Route 53 provide managed DNS services. Some DNS services can route traffic through various methods:

Disadvantage(s): DNS

Source(s) and further reading

Content delivery network

<p align="center"> <img src="http://i.imgur.com/h9TAuGI.jpg"/> <br/> <i><a href=https://www.creative-artworks.eu/why-use-a-content-delivery-network-cdn/>Source: Why use a CDN</a></i> </p>

A content delivery network (CDN) is a globally distributed network of proxy servers, serving content from locations closer to the user. Generally, static files such as HTML/CSS/JS, photos, and videos are served from CDN, although some CDNs such as Amazon's CloudFront support dynamic content. The site's DNS resolution will tell clients which server to contact.

Serving content from CDNs can significantly improve performance in two ways:

Push CDNs

Push CDNs receive new content whenever changes occur on your server. You take full responsibility for providing content, uploading directly to the CDN and rewriting URLs to point to the CDN. You can configure when content expires and when it is updated. Content is uploaded only when it is new or changed, minimizing traffic, but maximizing storage.

Sites with a small amount of traffic or sites with content that isn't often updated work well with push CDNs. Content is placed on the CDNs once, instead of being re-pulled at regular intervals.

Pull CDNs

Pull CDNs grab new content from your server when the first user requests the content. You leave the content on your server and rewrite URLs to point to the CDN. This results in a slower request until the content is cached on the CDN.

A time-to-live (TTL) determines how long content is cached. Pull CDNs minimize storage space on the CDN, but can create redundant traffic if files expire and are pulled before they have actually changed.

Sites with heavy traffic work well with pull CDNs, as traffic is spread out more evenly with only recently-requested content remaining on the CDN.

Disadvantage(s): CDN

Source(s) and further reading

Load balancer

<p align="center"> <img src="http://i.imgur.com/h81n9iK.png"/> <br/> <i><a href=http://horicky.blogspot.com/2010/10/scalable-system-design-patterns.html>Source: Scalable system design patterns</a></i> </p>

Load balancers distribute incoming client requests to computing resources such as application servers and databases. In each case, the load balancer returns the response from the computing resource to the appropriate client. Load balancers are effective at:

Load balancers can be implemented with hardware (expensive) or with software such as HAProxy.

Additional benefits include:

To protect against failures, it's common to set up multiple load balancers, either in active-passive or active-active mode.

Load balancers can route traffic based on various metrics, including:

Layer 4 load balancing

Layer 4 load balancers look at info at the transport layer to decide how to distribute requests. Generally, this involves the source, destination IP addresses, and ports in the header, but not the contents of the packet. Layer 4 load balancers forward network packets to and from the upstream server, performing Network Address Translation (NAT).

Layer 7 load balancing

Layer 7 load balancers look at the application layer to decide how to distribute requests. This can involve contents of the header, message, and cookies. Layer 7 load balancers terminates network traffic, reads the message, makes a load-balancing decision, then opens a connection to the selected server. For example, a layer 7 load balancer can direct video traffic to servers that host videos while directing more sensitive user billing traffic to security-hardened servers.

At the cost of flexibility, layer 4 load balancing requires less time and computing resources than Layer 7, although the performance impact can be minimal on modern commodity hardware.

Horizontal scaling

Load balancers can also help with horizontal scaling, improving performance and availability. Scaling out using commodity machines is more cost efficient and results in higher availability than scaling up a single server on more expensive hardware, called Vertical Scaling. It is also easier to hire for talent working on commodity hardware than it is for specialized enterprise systems.

Disadvantage(s): horizontal scaling

Disadvantage(s): load balancer

Source(s) and further reading

Reverse proxy (web server)

<p align="center"> <img src="http://i.imgur.com/n41Azff.png"/> <br/> <i><a href=https://upload.wikimedia.org/wikipedia/commons/6/67/Reverse_proxy_h2g2bob.svg>Source: Wikipedia</a></i> <br/> </p>

A reverse proxy is a web server that centralizes internal services and provides unified interfaces to the public. Requests from clients are forwarded to a server that can fulfill it before the reverse proxy returns the server's response to the client.

Additional benefits include:

Load balancer vs reverse proxy

Disadvantage(s): reverse proxy

Source(s) and further reading

Application layer

<p align="center"> <img src="http://i.imgur.com/yB5SYwm.png"/> <br/> <i><a href=http://lethain.com/introduction-to-architecting-systems-for-scale/#platform_layer>Source: Intro to architecting systems for scale</a></i> </p>

Separating out the web layer from the application layer (also known as platform layer) allows you to scale and configure both layers independently. Adding a new API results in adding application servers without necessarily adding additional web servers. The single responsibility principle advocates for small and autonomous services that work together. Small teams with small services can plan more aggressively for rapid growth.

Workers in the application layer also help enable asynchronism.

Microservices

Related to this discussion are microservices, which can be described as a suite of independently deployable, small, modular services. Each service runs a unique process and communicates through a well-defined, lightweight mechanism to serve a business goal. <sup><a href=https://smartbear.com/learn/api-design/what-are-microservices>1</a></sup>

Pinterest, for example, could have the following microservices: user profile, follower, feed, search, photo upload, etc.

Service Discovery

Systems such as Consul, Etcd, and Zookeeper can help services find each other by keeping track of registered names, addresses, and ports. Health checks help verify service integrity and are often done using an HTTP endpoint. Both Consul and Etcd have a built in key-value store that can be useful for storing config values and other shared data.

Disadvantage(s): application layer

Source(s) and further reading

Database

<p align="center"> <img src="http://i.imgur.com/Xkm5CXz.png"/> <br/> <i><a href=https://www.youtube.com/watch?v=w95murBkYmU>Source: Scaling up to your first 10 million users</a></i> </p>

Relational database management system (RDBMS)

A relational database like SQL is a collection of data items organized in tables.

ACID is a set of properties of relational database transactions.

There are many techniques to scale a relational database: master-slave replication, master-master replication, federation, sharding, denormalization, and SQL tuning.

Master-slave replication

The master serves reads and writes, replicating writes to one or more slaves, which serve only reads. Slaves can also replicate to additional slaves in a tree-like fashion. If the master goes offline, the system can continue to operate in read-only mode until a slave is promoted to a master or a new master is provisioned.

<p align="center"> <img src="http://i.imgur.com/C9ioGtn.png"/> <br/> <i><a href=http://www.slideshare.net/jboner/scalability-availability-stability-patterns/>Source: Scalability, availability, stability, patterns</a></i> </p>
Disadvantage(s): master-slave replication

Master-master replication

Both masters serve reads and writes and coordinate with each other on writes. If either master goes down, the system can continue to operate with both reads and writes.

<p align="center"> <img src="http://i.imgur.com/krAHLGg.png"/> <br/> <i><a href=http://www.slideshare.net/jboner/scalability-availability-stability-patterns/>Source: Scalability, availability, stability, patterns</a></i> </p>
Disadvantage(s): master-master replication
Disadvantage(s): replication
Source(s) and further reading: replication

Federation

<p align="center"> <img src="http://i.imgur.com/U3qV33e.png"/> <br/> <i><a href=https://www.youtube.com/watch?v=w95murBkYmU>Source: Scaling up to your first 10 million users</a></i> </p>

Federation (or functional partitioning) splits up databases by function. For example, instead of a single, monolithic database, you could have three databases: forums, users, and products, resulting in less read and write traffic to each database and therefore less replication lag. Smaller databases result in more data that can fit in memory, which in turn results in more cache hits due to improved cache locality. With no single central master serializing writes you can write in parallel, increasing throughput.

Disadvantage(s): federation
Source(s) and further reading: federation

Sharding

<p align="center"> <img src="http://i.imgur.com/wU8x5Id.png"/> <br/> <i><a href=http://www.slideshare.net/jboner/scalability-availability-stability-patterns/>Source: Scalability, availability, stability, patterns</a></i> </p>

Sharding distributes data across different databases such that each database can only manage a subset of the data. Taking a users database as an example, as the number of users increases, more shards are added to the cluster.

Similar to the advantages of federation, sharding results in less read and write traffic, less replication, and more cache hits. Index size is also reduced, which generally improves performance with faster queries. If one shard goes down, the other shards are still operational, although you'll want to add some form of replication to avoid data loss. Like federation, there is no single central master serializing writes, allowing you to write in parallel with increased throughput.

Common ways to shard a table of users is either through the user's last name initial or the user's geographic location.

Disadvantage(s): sharding
Source(s) and further reading: sharding

Denormalization

Denormalization attempts to improve read performance at the expense of some write performance. Redundant copies of the data are written in multiple tables to avoid expensive joins. Some RDBMS such as PostgreSQL and Oracle support materialized views which handle the work of storing redundant information and keeping redundant copies consistent.

Once data becomes distributed with techniques such as federation and sharding, managing joins across data centers further increases complexity. Denormalization might circumvent the need for such complex joins.

In most systems, reads can heavily outnumber writes 100:1 or even 1000:1. A read resulting in a complex database join can be very expensive, spending a significant amount of time on disk operations.

Disadvantage(s): denormalization
Source(s) and further reading: denormalization

SQL tuning

SQL tuning is a broad topic and many books have been written as reference.

It's important to benchmark and profile to simulate and uncover bottlenecks.

Benchmarking and profiling might point you to the following optimizations.

Tighten up the schema
Use good indices
Avoid expensive joins
Partition tables
Tune the query cache
Source(s) and further reading: SQL tuning

NoSQL

NoSQL is a collection of data items represented in a key-value store, document store, wide column store, or a graph database. Data is denormalized, and joins are generally done in the application code. Most NoSQL stores lack true ACID transactions and favor eventual consistency.

BASE is often used to describe the properties of NoSQL databases. In comparison with the CAP Theorem, BASE chooses availability over consistency.

In addition to choosing between SQL or NoSQL, it is helpful to understand which type of NoSQL database best fits your use case(s). We'll review key-value stores, document stores, wide column stores, and graph databases in the next section.

Key-value store

Abstraction: hash table

A key-value store generally allows for O(1) reads and writes and is often backed by memory or SSD. Data stores can maintain keys in lexicographic order, allowing efficient retrieval of key ranges. Key-value stores can allow for storing of metadata with a value.

Key-value stores provide high performance and are often used for simple data models or for rapidly-changing data, such as an in-memory cache layer. Since they offer only a limited set of operations, complexity is shifted to the application layer if additional operations are needed.

A key-value store is the basis for more complex systems such as a document store, and in some cases, a graph database.

Source(s) and further reading: key-value store

Document store

Abstraction: key-value store with documents stored as values

A document store is centered around documents (XML, JSON, binary, etc), where a document stores all information for a given object. Document stores provide APIs or a query language to query based on the internal structure of the document itself. Note, many key-value stores include features for working with a value's metadata, blurring the lines between these two storage types.

Based on the underlying implementation, documents are organized by collections, tags, metadata, or directories. Although documents can be organized or grouped together, documents may have fields that are completely different from each other.

Some document stores like MongoDB and CouchDB also provide a SQL-like language to perform complex queries. DynamoDB supports both key-values and documents.

Document stores provide high flexibility and are often used for working with occasionally changing data.

Source(s) and further reading: document store

Wide column store

<p align="center"> <img src="http://i.imgur.com/n16iOGk.png"/> <br/> <i><a href=http://blog.grio.com/2015/11/sql-nosql-a-brief-history.html>Source: SQL & NoSQL, a brief history</a></i> </p>

Abstraction: nested map ColumnFamily<RowKey, Columns<ColKey, Value, Timestamp>>

A wide column store's basic unit of data is a column (name/value pair). A column can be grouped in column families (analogous to a SQL table). Super column families further group column families. You can access each column independently with a row key, and columns with the same row key form a row. Each value contains a timestamp for versioning and for conflict resolution.

Google introduced Bigtable as the first wide column store, which influenced the open-source HBase often-used in the Hadoop ecosystem, and Cassandra from Facebook. Stores such as BigTable, HBase, and Cassandra maintain keys in lexicographic order, allowing efficient retrieval of selective key ranges.

Wide column stores offer high availability and high scalability. They are often used for very large data sets.

Source(s) and further reading: wide column store

Graph database

<p align="center"> <img src="http://i.imgur.com/fNcl65g.png"/> <br/> <i><a href=https://en.wikipedia.org/wiki/File:GraphDatabase_PropertyGraph.png>Source: Graph database</a></i> </p>

Abstraction: graph

In a graph database, each node is a record and each arc is a relationship between two nodes. Graph databases are optimized to represent complex relationships with many foreign keys or many-to-many relationships.

Graphs databases offer high performance for data models with complex relationships, such as a social network. They are relatively new and are not yet widely-used; it might be more difficult to find development tools and resources. Many graphs can only be accessed with REST APIs.

Source(s) and further reading: graph

Source(s) and further reading: NoSQL

SQL or NoSQL

<p align="center"> <img src="http://i.imgur.com/wXGqG5f.png"/> <br/> <i><a href=https://www.infoq.com/articles/Transition-RDBMS-NoSQL/>Source: Transitioning from RDBMS to NoSQL</a></i> </p>

Reasons for SQL:

Reasons for NoSQL:

Sample data well-suited for NoSQL:

Source(s) and further reading: SQL or NoSQL

Cache

<p align="center"> <img src="http://i.imgur.com/Q6z24La.png"/> <br/> <i><a href=http://horicky.blogspot.com/2010/10/scalable-system-design-patterns.html>Source: Scalable system design patterns</a></i> </p>

Caching improves page load times and can reduce the load on your servers and databases. In this model, the dispatcher will first lookup if the request has been made before and try to find the previous result to return, in order to save the actual execution.

Databases often benefit from a uniform distribution of reads and writes across its partitions. Popular items can skew the distribution, causing bottlenecks. Putting a cache in front of a database can help absorb uneven loads and spikes in traffic.

Client caching

Caches can be located on the client side (OS or browser), server side, or in a distinct cache layer.

CDN caching

CDNs are considered a type of cache.

Web server caching

Reverse proxies and caches such as Varnish can serve static and dynamic content directly. Web servers can also cache requests, returning responses without having to contact application servers.

Database caching

Your database usually includes some level of caching in a default configuration, optimized for a generic use case. Tweaking these settings for specific usage patterns can further boost performance.

Application caching

In-memory caches such as Memcached and Redis are key-value stores between your application and your data storage. Since the data is held in RAM, it is much faster than typical databases where data is stored on disk. RAM is more limited than disk, so cache invalidation algorithms such as least recently used (LRU) can help invalidate 'cold' entries and keep 'hot' data in RAM.

Redis has the following additional features:

There are multiple levels you can cache that fall into two general categories: database queries and objects:

Generally, you should try to avoid file-based caching, as it makes cloning and auto-scaling more difficult.

Caching at the database query level

Whenever you query the database, hash the query as a key and store the result to the cache. This approach suffers from expiration issues:

Caching at the object level

See your data as an object, similar to what you do with your application code. Have your application assemble the dataset from the database into a class instance or a data structure(s):

Suggestions of what to cache:

When to update the cache

Since you can only store a limited amount of data in cache, you'll need to determine which cache update strategy works best for your use case.

Cache-aside

<p align="center"> <img src="http://i.imgur.com/ONjORqk.png"/> <br/> <i><a href=http://www.slideshare.net/tmatyashovsky/from-cache-to-in-memory-data-grid-introduction-to-hazelcast>Source: From cache to in-memory data grid</a></i> </p>

The application is responsible for reading and writing from storage. The cache does not interact with storage directly. The application does the following:

def get_user(self, user_id):
    user = cache.get("user.{0}", user_id)
    if user is None:
        user = db.query("SELECT * FROM users WHERE user_id = {0}", user_id)
        if user is not None:
            key = "user.{0}".format(user_id)
            cache.set(key, json.dumps(user))
    return user

Memcached is generally used in this manner.

Subsequent reads of data added to cache are fast. Cache-aside is also referred to as lazy loading. Only requested data is cached, which avoids filling up the cache with data that isn't requested.

Disadvantage(s): cache-aside

Write-through

<p align="center"> <img src="http://i.imgur.com/0vBc0hN.png"/> <br/> <i><a href=http://www.slideshare.net/jboner/scalability-availability-stability-patterns/>Source: Scalability, availability, stability, patterns</a></i> </p>

The application uses the cache as the main data store, reading and writing data to it, while the cache is responsible for reading and writing to the database:

Application code:

set_user(12345, {"foo":"bar"})

Cache code:

def set_user(user_id, values):
    user = db.query("UPDATE Users WHERE id = {0}", user_id, values)
    cache.set(user_id, user)

Write-through is a slow overall operation due to the write operation, but subsequent reads of just written data are fast. Users are generally more tolerant of latency when updating data than reading data. Data in the cache is not stale.

Disadvantage(s): write through

Write-behind (write-back)

<p align="center"> <img src="http://i.imgur.com/rgSrvjG.png"/> <br/> <i><a href=http://www.slideshare.net/jboner/scalability-availability-stability-patterns/>Source: Scalability, availability, stability, patterns</a></i> </p>

In write-behind, the application does the following:

Disadvantage(s): write-behind

Refresh-ahead

<p align="center"> <img src="http://i.imgur.com/kxtjqgE.png"/> <br/> <i><a href=http://www.slideshare.net/tmatyashovsky/from-cache-to-in-memory-data-grid-introduction-to-hazelcast>Source: From cache to in-memory data grid</a></i> </p>

You can configure the cache to automatically refresh any recently accessed cache entry prior to its expiration.

Refresh-ahead can result in reduced latency vs read-through if the cache can accurately predict which items are likely to be needed in the future.

Disadvantage(s): refresh-ahead

Disadvantage(s): cache

Source(s) and further reading

Asynchronism

<p align="center"> <img src="http://i.imgur.com/54GYsSx.png"/> <br/> <i><a href=http://lethain.com/introduction-to-architecting-systems-for-scale/#platform_layer>Source: Intro to architecting systems for scale</a></i> </p>

Asynchronous workflows help reduce request times for expensive operations that would otherwise be performed in-line. They can also help by doing time-consuming work in advance, such as periodic aggregation of data.

Message queues

Message queues receive, hold, and deliver messages. If an operation is too slow to perform inline, you can use a message queue with the following workflow:

The user is not blocked and the job is processed in the background. During this time, the client might optionally do a small amount of processing to make it seem like the task has completed. For example, if posting a tweet, the tweet could be instantly posted to your timeline, but it could take some time before your tweet is actually delivered to all of your followers.

Redis is useful as a simple message broker but messages can be lost.

RabbitMQ is popular but requires you to adapt to the 'AMQP' protocol and manage your own nodes.

Amazon SQS is hosted but can have high latency and has the possibility of messages being delivered twice.

Task queues

Tasks queues receive tasks and their related data, runs them, then delivers their results. They can support scheduling and can be used to run computationally-intensive jobs in the background.

Celery has support for scheduling and primarily has python support.

Back pressure

If queues start to grow significantly, the queue size can become larger than memory, resulting in cache misses, disk reads, and even slower performance. Back pressure can help by limiting the queue size, thereby maintaining a high throughput rate and good response times for jobs already in the queue. Once the queue fills up, clients get a server busy or HTTP 503 status code to try again later. Clients can retry the request at a later time, perhaps with exponential backoff.

Disadvantage(s): asynchronism

Source(s) and further reading

Communication

<p align="center"> <img src="http://i.imgur.com/5KeocQs.jpg"/> <br/> <i><a href=http://www.escotal.com/osilayer.html>Source: OSI 7 layer model</a></i> </p>

Hypertext transfer protocol (HTTP)

HTTP is a method for encoding and transporting data between a client and a server. It is a request/response protocol: clients issue requests and servers issue responses with relevant content and completion status info about the request. HTTP is self-contained, allowing requests and responses to flow through many intermediate routers and servers that perform load balancing, caching, encryption, and compression.

A basic HTTP request consists of a verb (method) and a resource (endpoint). Below are common HTTP verbs:

VerbDescriptionIdempotent*SafeCacheable
GETReads a resourceYesYesYes
POSTCreates a resource or trigger a process that handles dataNoNoYes if response contains freshness info
PUTCreates or replace a resourceYesNoNo
PATCHPartially updates a resourceNoNoYes if response contains freshness info
DELETEDeletes a resourceYesNoNo

*Can be called many times without different outcomes.

HTTP is an application layer protocol relying on lower-level protocols such as TCP and UDP.

Source(s) and further reading: HTTP

Transmission control protocol (TCP)

<p align="center"> <img src="http://i.imgur.com/JdAsdvG.jpg"/> <br/> <i><a href=http://www.wildbunny.co.uk/blog/2012/10/09/how-to-make-a-multi-player-game-part-1/>Source: How to make a multiplayer game</a></i> </p>

TCP is a connection-oriented protocol over an IP network. Connection is established and terminated using a handshake. All packets sent are guaranteed to reach the destination in the original order and without corruption through:

If the sender does not receive a correct response, it will resend the packets. If there are multiple timeouts, the connection is dropped. TCP also implements flow control and congestion control. These guarantees cause delays and generally result in less efficient transmission than UDP.

To ensure high throughput, web servers can keep a large number of TCP connections open, resulting in high memory usage. It can be expensive to have a large number of open connections between web server threads and say, a memcached server. Connection pooling can help in addition to switching to UDP where applicable.

TCP is useful for applications that require high reliability but are less time critical. Some examples include web servers, database info, SMTP, FTP, and SSH.

Use TCP over UDP when:

User datagram protocol (UDP)

<p align="center"> <img src="http://i.imgur.com/yzDrJtA.jpg"/> <br/> <i><a href=http://www.wildbunny.co.uk/blog/2012/10/09/how-to-make-a-multi-player-game-part-1/>Source: How to make a multiplayer game</a></i> </p>

UDP is connectionless. Datagrams (analogous to packets) are guaranteed only at the datagram level. Datagrams might reach their destination out of order or not at all. UDP does not support congestion control. Without the guarantees that TCP support, UDP is generally more efficient.

UDP can broadcast, sending datagrams to all devices on the subnet. This is useful with DHCP because the client has not yet received an IP address, thus preventing a way for TCP to stream without the IP address.

UDP is less reliable but works well in real time use cases such as VoIP, video chat, streaming, and realtime multiplayer games.

Use UDP over TCP when:

Source(s) and further reading: TCP and UDP

Remote procedure call (RPC)

<p align="center"> <img src="http://i.imgur.com/iF4Mkb5.png"/> <br/> <i><a href=http://www.puncsky.com/blog/2016-02-13-crack-the-system-design-interview>Source: Crack the system design interview</a></i> </p>

In an RPC, a client causes a procedure to execute on a different address space, usually a remote server. The procedure is coded as if it were a local procedure call, abstracting away the details of how to communicate with the server from the client program. Remote calls are usually slower and less reliable than local calls so it is helpful to distinguish RPC calls from local calls. Popular RPC frameworks include Protobuf, Thrift, and Avro.

RPC is a request-response protocol:

Sample RPC calls:

GET /someoperation?data=anId

POST /anotheroperation
{
  "data":"anId";
  "anotherdata": "another value"
}

RPC is focused on exposing behaviors. RPCs are often used for performance reasons with internal communications, as you can hand-craft native calls to better fit your use cases.

Choose a native library (aka SDK) when:

HTTP APIs following REST tend to be used more often for public APIs.

Disadvantage(s): RPC

Representational state transfer (REST)

REST is an architectural style enforcing a client/server model where the client acts on a set of resources managed by the server. The server provides a representation of resources and actions that can either manipulate or get a new representation of resources. All communication must be stateless and cacheable.

There are four qualities of a RESTful interface:

Sample REST calls:

GET /someresources/anId

PUT /someresources/anId
{"anotherdata": "another value"}

REST is focused on exposing data. It minimizes the coupling between client/server and is often used for public HTTP APIs. REST uses a more generic and uniform method of exposing resources through URIs, representation through headers, and actions through verbs such as GET, POST, PUT, DELETE, and PATCH. Being stateless, REST is great for horizontal scaling and partitioning.

Disadvantage(s): REST

RPC and REST calls comparison

OperationRPCREST
SignupPOST /signupPOST /persons
ResignPOST /resign<br/>{<br/>"personid": "1234"<br/>}DELETE /persons/1234
Read a personGET /readPerson?personid=1234GET /persons/1234
Read a person’s items listGET /readUsersItemsList?personid=1234GET /persons/1234/items
Add an item to a person’s itemsPOST /addItemToUsersItemsList<br/>{<br/>"personid": "1234";<br/>"itemid": "456"<br/>}POST /persons/1234/items<br/>{<br/>"itemid": "456"<br/>}
Update an itemPOST /modifyItem<br/>{<br/>"itemid": "456";<br/>"key": "value"<br/>}PUT /items/456<br/>{<br/>"key": "value"<br/>}
Delete an itemPOST /removeItem<br/>{<br/>"itemid": "456"<br/>}DELETE /items/456
<p align="center"> <i><a href=https://apihandyman.io/do-you-really-know-why-you-prefer-rest-over-rpc/>Source: Do you really know why you prefer REST over RPC</a></i> </p>

Source(s) and further reading: REST and RPC

Security

This section could use some updates. Consider contributing!

Security is a broad topic. Unless you have considerable experience, a security background, or are applying for a position that requires knowledge of security, you probably won't need to know more than the basics:

Source(s) and further reading

Appendix

You'll sometimes be asked to do 'back-of-the-envelope' estimates. For example, you might need to determine how long it will take to generate 100 image thumbnails from disk or how much memory a data structure will take. The Powers of two table and Latency numbers every programmer should know are handy references.

Powers of two table

Power           Exact Value         Approx Value        Bytes
---------------------------------------------------------------
7                             128
8                             256
10                           1024   1 thousand           1 KB
16                         65,536                       64 KB
20                      1,048,576   1 million            1 MB
30                  1,073,741,824   1 billion            1 GB
32                  4,294,967,296                        4 GB
40              1,099,511,627,776   1 trillion           1 TB

Source(s) and further reading

Latency numbers every programmer should know

Latency Comparison Numbers
--------------------------
L1 cache reference                           0.5 ns
Branch mispredict                            5   ns
L2 cache reference                           7   ns                      14x L1 cache
Mutex lock/unlock                           25   ns
Main memory reference                      100   ns                      20x L2 cache, 200x L1 cache
Compress 1K bytes with Zippy            10,000   ns       10 us
Send 1 KB bytes over 1 Gbps network     10,000   ns       10 us
Read 4 KB randomly from SSD*           150,000   ns      150 us          ~1GB/sec SSD
Read 1 MB sequentially from memory     250,000   ns      250 us
Round trip within same datacenter      500,000   ns      500 us
Read 1 MB sequentially from SSD*     1,000,000   ns    1,000 us    1 ms  ~1GB/sec SSD, 4X memory
Disk seek                           10,000,000   ns   10,000 us   10 ms  20x datacenter roundtrip
Read 1 MB sequentially from 1 Gbps  10,000,000   ns   10,000 us   10 ms  40x memory, 10X SSD
Read 1 MB sequentially from disk    30,000,000   ns   30,000 us   30 ms 120x memory, 30X SSD
Send packet CA->Netherlands->CA    150,000,000   ns  150,000 us  150 ms

Notes
-----
1 ns = 10^-9 seconds
1 us = 10^-6 seconds = 1,000 ns
1 ms = 10^-3 seconds = 1,000 us = 1,000,000 ns

Handy metrics based on numbers above:

Latency numbers visualized

Source(s) and further reading

Additional system design interview questions

Common system design interview questions, with links to resources on how to solve each.

QuestionReference(s)
Design a file sync service like Dropboxyoutube.com
Design a search engine like Googlequeue.acm.org<br/>stackexchange.com<br/>ardendertat.com<br/>stanford.edu
Design a scalable web crawler like Googlequora.com
Design Google docscode.google.com<br/>neil.fraser.name
Design a key-value store like Redisslideshare.net
Design a cache system like Memcachedslideshare.net
Design a recommendation system like Amazon'shulu.com<br/>ijcai13.org
Design a tinyurl system like Bitlyn00tc0d3r.blogspot.com
Design a chat app like WhatsApphighscalability.com
Design a picture sharing system like Instagramhighscalability.com<br/>highscalability.com
Design the Facebook news feed functionquora.com<br/>quora.com<br/>slideshare.net
Design the Facebook timeline functionfacebook.com<br/>highscalability.com
Design the Facebook chat functionerlang-factory.com<br/>facebook.com
Design a graph search function like Facebook'sfacebook.com<br/>facebook.com<br/>facebook.com
Design a content delivery network like CloudFlarefigshare.com
Design a trending topic system like Twitter'smichael-noll.com<br/>snikolov .wordpress.com
Design a random ID generation systemblog.twitter.com<br/>github.com
Return the top k requests during a time intervalcs.ucsb.edu<br/>wpi.edu
Design a system that serves data from multiple data centershighscalability.com
Design an online multiplayer card gameindieflashblog.com<br/>buildnewgames.com
Design a garbage collection systemstuffwithstuff.com<br/>washington.edu
Design an API rate limiterhttps://stripe.com/blog/
Add a system design questionContribute

Real world architectures

Articles on how real world systems are designed.

<p align="center"> <img src="http://i.imgur.com/TcUo2fw.png"/> <br/> <i><a href=https://www.infoq.com/presentations/Twitter-Timeline-Scalability>Source: Twitter timelines at scale</a></i> </p>

Don't focus on nitty gritty details for the following articles, instead:

TypeSystemReference(s)
Data processingMapReduce - Distributed data processing from Googleresearch.google.com
Data processingSpark - Distributed data processing from Databricksslideshare.net
Data processingStorm - Distributed data processing from Twitterslideshare.net
Data storeBigtable - Distributed column-oriented database from Googleharvard.edu
Data storeHBase - Open source implementation of Bigtableslideshare.net
Data storeCassandra - Distributed column-oriented database from Facebookslideshare.net
Data storeDynamoDB - Document-oriented database from Amazonharvard.edu
Data storeMongoDB - Document-oriented databaseslideshare.net
Data storeSpanner - Globally-distributed database from Googleresearch.google.com
Data storeMemcached - Distributed memory caching systemslideshare.net
Data storeRedis - Distributed memory caching system with persistence and value typesslideshare.net
File systemGoogle File System (GFS) - Distributed file systemresearch.google.com
File systemHadoop File System (HDFS) - Open source implementation of GFSapache.org
MiscChubby - Lock service for loosely-coupled distributed systems from Googleresearch.google.com
MiscDapper - Distributed systems tracing infrastructureresearch.google.com
MiscKafka - Pub/sub message queue from LinkedInslideshare.net
MiscZookeeper - Centralized infrastructure and services enabling synchronizationslideshare.net
Add an architectureContribute

Company architectures

CompanyReference(s)
AmazonAmazon architecture
CinchcastProducing 1,500 hours of audio every day
DataSiftRealtime datamining At 120,000 tweets per second
DropBoxHow we've scaled Dropbox
ESPNOperating At 100,000 duh nuh nuhs per second
GoogleGoogle architecture
Instagram14 million users, terabytes of photos<br/>What powers Instagram
Justin.tvJustin.Tv's live video broadcasting architecture
FacebookScaling memcached at Facebook<br/>TAO: Facebook’s distributed data store for the social graph<br/>Facebook’s photo storage<br/>How Facebook Live Streams To 800,000 Simultaneous Viewers
FlickrFlickr architecture
MailboxFrom 0 to one million users in 6 weeks
NetflixA 360 Degree View Of The Entire Netflix Stack<br/>Netflix: What Happens When You Press Play?
PinterestFrom 0 To 10s of billions of page views a month<br/>18 million visitors, 10x growth, 12 employees
Playfish50 million monthly users and growing
PlentyOfFishPlentyOfFish architecture
SalesforceHow they handle 1.3 billion transactions a day
Stack OverflowStack Overflow architecture
TripAdvisor40M visitors, 200M dynamic page views, 30TB data
Tumblr15 billion page views a month
TwitterMaking Twitter 10000 percent faster<br/>Storing 250 million tweets a day using MySQL<br/>150M active users, 300K QPS, a 22 MB/S firehose<br/>Timelines at scale<br/>Big and small data at Twitter<br/>Operations at Twitter: scaling beyond 100 million users<br/>How Twitter Handles 3,000 Images Per Second
UberHow Uber scales their real-time market platform<br/>Lessons Learned From Scaling Uber To 2000 Engineers, 1000 Services, And 8000 Git Repositories
WhatsAppThe WhatsApp architecture Facebook bought for $19 billion
YouTubeYouTube scalability<br/>YouTube architecture

Company engineering blogs

Architectures for companies you are interviewing with.

Questions you encounter might be from the same domain.

Source(s) and further reading

Looking to add a blog? To avoid duplicating work, consider adding your company blog to the following repo:

Under development

Interested in adding a section or helping complete one in-progress? Contribute!

Credits

Credits and sources are provided throughout this repo.

Special thanks to:

Contact info

Feel free to contact me to discuss any issues, questions, or comments.

My contact info can be found on my GitHub page.

License

I am providing code and resources in this repository to you under an open source license. Because this is my personal repository, the license you receive to my code and resources is from me and not my employer (Facebook).

Copyright 2017 Donne Martin

Creative Commons Attribution 4.0 International License (CC BY 4.0)

http://creativecommons.org/licenses/by/4.0/