How many socket connections can one server handle? | InterviewReady posted on the topic | LinkedIn (2024)

Hey all! Do you have any queries about system design?If so, don't forget to join the Zoom class tomorrow!To add all of our Zoom classes to your schedule, integrate your Google Calendar. Over 20,000 engineers have joined us to learn about System Design. We are glad to be a part of your journey.Here is the link: https://lnkd.in/dvcAbyWDSee you tomorrow!#systemdesign #whitepaper #database

9

Why are Content Delivery Networks (CDN) so popular?Delivering content across the globe quickly is a huge challenge. Since low latency is an indicator of high quality (Amazon and Google studies show a direct correlation between website latency and sales), developers are constantly looking to make their web pages faster. Content Delivery Networks (CDNs) are vital to this effort.1️⃣ CDNs are servers deployed across the globe, allowing users to connect to nearby servers.2️⃣ This improves availability, and spreads request load, reducing single points of failure.3️⃣ The shorter round trips and reduced bandwidth usage translate to reduced tech costs.4️⃣ Websites can use these systems as a black box, with negligible maintainence and setup costs.Key components:1. Origin Server: The principal source of content.2. Edge servers: Caches dispersed to provide content.3. DNS: Directs requests to the closest edge server.4. Control Plane: Manages CDN infrastructure.Popular CDN providers:1. Cloudflare2. Akamai3. Amazon CloudFront4. FastlyUseful tips:1. Use appropriate TTLs, else content updates result in restarts :p 2. CDNs are great for DDoS protection. 3. You can set content-specific optimizations in them, like image compressions.4. Choose a suitable caching strategy for the CDN servers.Have you used CDNs in your projects, and if so how did you keep them up to date?#SystemDesign #CDN #Interviewready

10

This is a must read if you are a senior software or data engineer.The Google Dremel Paper explains how to efficiently run ad-hoc queries. This 10-page article lists the challenges of running queries on huge datasets, along with amazing solutions.Key points:1. Running aggregate queries like count, max, and min quickly on huge data sets is a challenge. For example, “How many users from Bangalore frequently move and search for cars?”2. Dremel runs on thousands of nodes for parallelism. 3. It uses a multi-level serving tree, inspired by distributed search engines, to run parallel queries.4. It has a query optimizer, similar to those in SQL databases but optimized for distributed execution.5. Dremel represents data in columnar form to allow rapid scans.6. It uses algorithms like Finite Automata and modified Segment Trees.What makes Dremel amazing:1. Dremel was awarded as the VLDB 2020 Test of Time winner. Systems like Apache Impala have been designed based on Dremel.2. Dremel allows interactive queries on petabyte-sized datasets efficiently.3. It handles deeply nested data and executes massively distributed queries.To learn more about software engineering and system design, check out our course: https://interviewready.io/#SystemDesign #Google #Engineering

7

Load balancers are like traffic cops on the internet 🚦Here's how they work:1️⃣ Clients send requests to a gateway (a server accepting external requests).2️⃣ The gateway talks to the load balancer, which selects the best machine to serve this request.3️⃣ The gateway then forwards this request to the assigned server.4️⃣ Everyone is happy.That's it. The benefits of this approach are:1. Scalability: Servers can be added/removed efficiently2. High availability: Traffic is re-directed from failed servers to healthy servers3. Performance: Work is evenly distributed for smaller worst-case request queue size (improving worst-case response times)Commonly employed algorithms include:1. Round Robin: Go through servers one by one using an iterator.2. Least Connections: Send requests to the server having the fewest active connections.3. IP Hash: Use the client’s IP to select a server for session persistence (caching is awesome).Useful tips: 1. Load balancing can be on layer 4 (transport level) and layer 7 (application level).2. Regularly check the health of servers to decide on adding or removing a few.3. SSL termination helps offload encryption headaches.Have you used a load balancer in your system, and if so what algorithm did you use?Watch this video on load balancer here:Link: https://lnkd.in/dtqhwAsd#LoadBalancer #SystemDesign #Developers

47

1 Comment

We wish you a very Happy Ganesh Chaturthi!May Bappa bless you with wisdom, prosperity and happiness on this auspicious occasion. 🌺Ganpati Bappa Morya! 🙏#ganeshchaturthi #india #festival

11

This case study shows how Canva collects 25 Billion events per day.Challenges:1. Canva has more than 175 million monthly active-users2. Users connect on different platforms like iOS, android, web, and desktop apps3. More than 25 Billion events are generated per day4. SLAs of 99.999% uptime.Canva's Solution:Canva first tried using AWS SQS and SNS for event ingestion.After examining these options, they decided to go with AWS Kinesis Data Streams.There are a few reasons Kinesis made sense:1. Significantly lower cost of operating (85% less compared to SQS/SNS)2. Real-time data streaming with sub-second latency3. Scalable (necessary to handle massive event volumes)4. Seamless integration with other Amazon Web Services (AWS) services.5. Data can be retained and replayed. This is a standard Event Sourcing benefit.Results:1. Canva now manages 25 billion events per day.2. The cost of their Kinesis solution is 85% less than that of running SQS/SNS.3. Annual savings of $600k due to event compression.4. Meeting the requirement of 99.999% uptime as requested.Key Takeaways:1. The right data streaming service can dramatically impact costs at scale. The scale affects pricing significantly.2. Batching and compression strategies can lead to significant savings. Memory and bandwidth are saved through batching and compression.3. Decoupling ingestion from processing helps with scalability. Think of the CQRS pattern.4. A hybrid approach (Kinesis + SQS) can address different processing needs effectively. This allows Canva to serve a wide range of data analytics use cases.This case study shows that choosing the right tech, with practical improvements like compression, can massively reduce costs. Did you face similar challenges when building a data pipeline?Let me know in the comments!#Canva #SystemDesign #CaseStudy

99

4 Comments

"Cache me if you can" 🏃 But what is CACHING?Caching is the practice of copying frequently accessed data into memory. Subsequent requests for the same data are served quickly from memory, instead of a slow database. The goal of caching is to reduce latency by avoiding database calls.Here are 4 commonly used caching techniques:🔷 Write-Behind Pattern:Writes are immediately acknowledged in the cache and asynchronously updated in the database. This reduces write latency and handles high write loads efficiently, but risks data loss if the system fails before persisting to the database. Ideal for scenarios prioritizing write performance over immediate consistency.🔷 Refresh-Ahead Pattern:Proactively updates cached data before expiration based on predicted access patterns. This reduces read latency for frequently accessed data but may waste resources updating unused data. Best for systems with predictable access patterns requiring minimal read latency.🔷 Write-Through Pattern:Simultaneously writes data to both cache and database in a single transaction. Ensures strong data consistency but increases write latency. Suitable for applications requiring data integrity and consistency, tolerating slower write operations.🔷 Write-Around Pattern:Bypasses the cache for writes, updating only the database. The cache is populated on read operations. This prevents cache pollution from write-heavy workloads but can increase read latency for recently written data. Effective for systems with high write volumes and infrequent reads of new data.The right caching strategy depends on system requirements and constraints. It's often useful to think of a hybrid caching strategy based on system tradeoffs.-------Checkout our website & see how you can upskill: https://interviewready.io/#caching #systemdesign #interview

146

4 Comments

Some low-level design patterns for senior engineers:1. Read your writes: https://lnkd.in/dJYZ3Z_72. Asynchronous Processing with Futures: https://lnkd.in/d55-j-G83. Request Collapsing: https://lnkd.in/dFbJmwJC4. What are threads: https://lnkd.in/dF4q9ez25. Ordering requests with threads: https://lnkd.in/dV_fqwW86. Event Bus Publishers and Subscribers: https://lnkd.in/dcciJ4vM7. What is a software client library: https://lnkd.in/dbYzNXeH8. Classes and package structuring: https://lnkd.in/dqagPDG99. Dependency Inversion: https://lnkd.in/djU2uw4K10. What is backpressure: https://lnkd.in/dksjQb7NIf you work as a software engineer, you are probably using a few of these patterns already :DCheers!#SoftwareEngineering #LowLevelDesign #DesignPatterns

28

Last one day left before we close the offer & thank you for all the love and enthusiasm you all have shown!For everyone who has registered for the course, we cannot wait to meet you all in the live class ❤️ and for those who missed the sale we have good news for you -Last ONE DAY before we close the 70% offer on our system design course.Take a step towards your dream job and grab your course NOW 👇https://interviewready.io/Code: DESIGN70#sale #discount #systemdesign

5

System design of Zerodha is a popular interview question asked in tech companies.The idea is to see how candidates design a real-time large-scale distributed system that can accept buy and sell orders.In this design question, you are asked to place the missing components in the marked boxes. We also have a capacity estimation question to test your math skills!Design Link: https://lnkd.in/dF-9Tb4RThe month-end sale is live - Get 70% off on the System Design course!Buy course: http://interviewready.io/

8