What is CAP theorem?
CAP(Consistency-Availability-Partition Tolerance) theorem says that a distributed system cannot guarantee C, A and P simultaneously. It can at max provide any 2 of the 3 guarantees. Let us understand this with the help of a distributed database system.
- Consistency: This states that the data has to remain consistent after the execution of an operation in the database. For example, post database updating, all queries should retrieve the same result. - Availability: The databases cannot have downtime and should be available and responsive always. - Partition Tolerance: The database system should be functioning despite the communication becoming unstable.
How is Horizontal scaling different from Vertical scaling?
- Horizontal scaling refers to the addition of more computing machines to the network that shares the processing and memory workload across a distributed network of devices. In simple words, more instances of servers are added to the existing pool and the traffic load is distributed across these devices in an efficient manner.
- Vertical scaling refers to the concept of upgrading the resource capacity such as increasing RAM, adding efficient processors etc of a single machine or switching to a new machine with more capacity. The capability of the server can be enhanced without the need for code manipulation.
What do you understand by load balancing? Why is it important in system design?
Load balancing refers to the concept of distributing incoming traffic efficiently across a group of various backend servers. These servers are called server pools. Modern-day websites are designed to serve millions of requests from clients and return the responses in a fast and reliable manner. In order to serve these requests, the addition of more servers is required. In such a scenario, it is essential to distribute request traffic efficiently across each server so that they do not face undue loads. Load balancer acts as a traffic police cop facing the requests and routes them across the available servers in a way that not a single server is overwhelmed which could possibly degrade the application performance.
When a server goes down, the load balancer redirects traffic to the remaining available servers. When a new server gets added to the configuration, the requests are automatically redirected to it. Following are the benefits of load balancers:
- They help to prevent requests from going to unhealthy or unavailable servers. - Helps to prevent resources overloading. - Helps to eliminate a single point of failure since the requests are routed to available servers whenever a server goes down. - Requests sent to the servers are encrypted and the responses are decrypted. It aids in SSL termination and removes the need to install X.509 certificates on every server. - Load balancing impacts system security and allows continuous software updates for accommodating changes in the system.
What do you understand by Latency, throughput, and availability of a system?
Performance is an important factor in system design as it helps in making our services fast and reliable. Following are the three key metrics for measuring the performance:
- Latency: This is the time taken in milliseconds for delivering a single message. - Throughput: This is the amount of data successfully transmitted through a system in a given amount of time. It is measured in bits per second. - Availability: This determines the amount of time a system is available to respond to requests. It is calculated: System Uptime / (System Uptime+Downtime).
What is Sharding?
Sharding is a process of splitting the large logical dataset into multiple databases. It also refers to horizontal partitioning of data as it will be stored on multiple machines. By doing so, a sharded database becomes capable of handling more requests than a single large machine. Consider an example - in the following image, assume that we have around 1TB of data present in the database, when we perform sharding, we divide the large 1TB data into smaller chunks of 256GB into partitions called shards.
Sharding helps to scale databases by helping to handle the increased load by providing increased throughput, storage capacity and ensuring high availability.
How is sharding different from partitioning?
- Database Sharding - Sharding is a technique for dividing a single dataset among many databases, allowing it to be stored across multiple workstations. Larger datasets can be divided into smaller parts and stored in numerous data nodes, boosting the system’s total storage capacity. A sharded database, similarly, can accommodate more requests than a single system by dividing the data over numerous machines. Sharding, also known as horizontal scaling or scale-out, is a type of scaling in which more nodes are added to distribute the load. Horizontal scaling provides near-limitless scalability for handling large amounts of data and high-volume tasks.
- Database Partitioning - Partitioning is the process of separating stored database objects (tables, indexes, and views) into distinct portions. Large database items are partitioned to improve controllability, performance, and availability. Partitioning can enhance performance when accessing partitioned tables in specific instances. Partitioning can act as a leading column in indexes, reducing index size and increasing the likelihood of finding the most desired indexes in memory. When a large portion of one area is used in the resultset, scanning that region is much faster than accessing data scattered throughout the entire table by index. Adding and deleting sections allows for large-scale data uploading and deletion, which improves performance. Data that are rarely used can be uploaded to more affordable data storage devices.
How is performance and scalability related to each other?
A system is said to be scalable if there is increased performance is proportional to the resources added. Generally, performance increase in terms of scalability refers to serving more work units. But this can also mean being able to handle larger work units when datasets grow. If there is a performance problem in the application, then the system will be slow only for a single user. But if there is a scalability problem, then the system may be fast for a single user but it can get slow under heavy user load on the application.
What are the various Consistency patterns available in system design?
Consistency from the CAP theorem states that every read request should get the most recently written data. When there are multiple data copies available, there arises a problem of synchronizing them so that the clients get fresh data consistently. Following are the consistency patterns available:
- Weak consistency: After a data write, the read request may or may not be able to get the new data. This type of consistency works well in real-time use cases like VoIP, video chat, real-time multiplayer games etc. For example, when we are on a phone call, if we lose network for a few seconds, then we lose information about what was spoken during that time. - Eventual consistency: Post data write, the reads will eventually see the latest data within milliseconds. Here, the data is replicated asynchronously. These are seen in DNS and email systems. This works well in highly available systems. - Strong consistency: After a data write, the subsequent reads will see the latest data. Here, the data is replicated synchronously. This is seen in RDBMS and file systems and are suitable in systems requiring transactions of data.
What do you understand by Content delivery network?
Content delivery network or in short CDN is a globally distributed proxy server network that serves content from locations close by to the end-users. Usually, in websites, static files like HTML, CSS, JS files, images and videos are served from CDN.
There are two types of CDNs, they are:
- Push CDNs: Here, the content is received by the CDNs whenever changes occur on the server. The responsibility lies in us for uploading the content to CDNs. Content gets updated to the CDN only when it is modified or added which in turn maximises storage by minimising the traffic. Generally, sites with lesser traffic or content work well using push CDNs. - Pull CDNs: Here new content is grabbed from the server when the first user requests the content from the site. This leads to slower requests for the first time till the content gets stored/cached on the CDN. These CDNs minimizes space utilized on CDN but can lead to redundant traffic when expired files are pulled before they are changed. Websites having heavy traffic work well when used with pull CDNs.
What do you understand by Leader Election?
In a distributed environment where there are multiple servers contributing to the availability of the application, there can be situations where only one server has to take lead for updating third party APIs as different servers could cause problems while using the third party APIs. This server is called the primary server and the process of choosing this server is called leader election. The servers in the distributed environment have to detect when the leader server has failed and appoint another one to become a leader. This process is most suitable in high availability and strong consistency based applications by using a consensus algorithm.
What are some of the design issues in distributed systems?
- Heterogeneity: The Internet allows applications to run over a heterogeneous collection of computers and networks. There would be different types of networks and the differences are masked by the usage of standard Internet protocols for communicating with each other. This becomes an issue while designing distributed applications
- Openness: Openness represents the measure by which a system can be extended and re-implemented in different ways. In distributed systems, it specifies the degree to which new sharing services can be added and made available for client usage.
- Security: The information maintained in distributed systems need to be secure as they are valuable to the users. The confidentiality, availability and integrity of the distributed systems have to be maintained and this sometimes becomes a challenge.
- Scalability: A system is scalable if it remains effective when there is a significant increase in the request traffic and resources. Designing a distributed system involves planning well in advance how well the system can be made scalable under varying user loads.
- Failure Handling: In a distributed environment, the failures are partial, meaning if some components fail, others would still function. It becomes challenging to handle these failures as it involves identifying right components where the failures occur.
