Archives
-
Un enfoque proactivo para navegar por la geopolítica
by "Destacados de McKinsey" <publishing@email.mckinsey.com> - 07:34 - 14 Dec 2024 -
The year's most popular articles from the McKinsey Quarterly
McKinsey&Company
Share this email
As 2024 draws to a close, we’re revisiting our most popular insights of the year. Today, we give you our best-read pieces from the McKinsey Quarterly, which focus on topics including the human side of gen AI, rethinking your talent strategy, and the latest on healthy organizations. At No. 1 is "A generative AI reset: Rewiring to turn potential into value in 2024," by McKinsey’s Eric Lamarre, Alex Singla, Alexander Sukharevsky, and Rodney Zemmel. Read on for our full top 10, and don't miss McKinsey Publishing's "Meet the moment"—our best insights from 2024 to help you navigate change.
This email contains information about McKinsey’s research, insights, services, or events. By opening our emails or clicking on links, you agree to our use of cookies and web tracking technology. For more information on how we use and protect your information, please review our privacy policy.
You received this email because you subscribed to our McKinsey Quarterly alert list.
Copyright © 2024 | McKinsey & Company, 3 World Trade Center, 175 Greenwich Street, New York, NY 10007
by "McKinsey Top Ten" <publishing@email.mckinsey.com> - 09:37 - 13 Dec 2024 -
When tailoring stakeholder comms, think EDGE
The Shortlist
Share this email
Curated by Alex Panas, global leader of industries, & Axel Karlsson, global leader of functional practices and growth platforms
Welcome to the latest edition of the CEO Shortlist, a biweekly newsletter of our best ideas for the C-suite. This week, we feature our latest thinking on how CEOs can engage their stakeholders. We appreciate the opportunity to connect and hope you find our perspectives novel and insightful. Please let us know what you think! You can reach us at Alex_Panas@mckinsey.com and Axel_Karlsson@mckinsey.com. Thank you.
—Alex and Axel
Shareholders. Employees. Customers and governments. All are critical constituents for big-league CEOs. And according to our research, the world’s best CEOs know how to communicate clearly with these disparate audiences. We’ve distilled this level of engagement into four key elements. Leading CEOs have an expanded view of their role as the company’s bridge to the external world, a distinctive narrative as “storyteller in chief,” a growth-oriented mindset that empowers a team of internal and external ambassadors to articulate the company’s vision, and an engaged posture that systematically strengthens stakeholder connections. Read our latest insights and create your own EDGE.
We hope you find these ideas inspiring and helpful. See you in 2025 with more McKinsey ideas for the CEO and others in the C-suite.Share these insights
This email contains information about McKinsey’s research, insights, services, or events. By opening our emails or clicking on links, you agree to our use of cookies and web tracking technology. For more information on how we use and protect your information, please review our privacy policy.
You received this email because you subscribed to The CEO Shortlist newsletter.
Copyright © 2024 | McKinsey & Company, 3 World Trade Center, 175 Greenwich Street, New York, NY 10007
by "McKinsey CEO Shortlist" <publishing@email.mckinsey.com> - 03:31 - 13 Dec 2024 -
How retailers are enticing shoppers this holiday season
Only McKinsey Perspectives
Brought to you by Alex Panas, global leader of industries, & Axel Karlsson, global leader of functional practices and growth platforms
Welcome to the latest edition of Only McKinsey Perspectives. We hope you find our insights useful. Let us know what you think at Alex_Panas@McKinsey.com and Axel_Karlsson@McKinsey.com.
—Alex and Axel
•
Looking to save. Heading into the year-end holidays, US consumers are not feeling good about their economic prospects. Consumer sentiment is lower than it was prepandemic, explains McKinsey partner Tamara Charm, and when it comes to holiday shopping in 2024, consumers are hoping to save money. Charm says that they’re looking for sales and will be shopping throughout the holiday season to try to get what they want for less.
—Edited by Belinda Yu, editor, Atlanta
This email contains information about McKinsey's research, insights, services, or events. By opening our emails or clicking on links, you agree to our use of cookies and web tracking technology. For more information on how we use and protect your information, please review our privacy policy.
You received this email because you subscribed to the Only McKinsey Perspectives newsletter, formerly known as Only McKinsey.
Copyright © 2024 | McKinsey & Company, 3 World Trade Center, 175 Greenwich Street, New York, NY 10007
by "Only McKinsey Perspectives" <publishing@email.mckinsey.com> - 12:43 - 13 Dec 2024 -
Database Performance Demystified: Essential Tips and Strategies
Database Performance Demystified: Essential Tips and Strategies
Forwarded this email? Subscribe here for moreLatest articles
If you’re not a subscriber, here’s what you missed this month.
Mastering Modern Authentication: Cookies, Sessions, JWT, and PASETO
Stateless Architecture: The Key to Building Scalable and Resilient Systems
Distributed Caching: The Secret to High-Performance Applications
To receive all the full articles and support ByteByteGo, consider subscribing:
Databases are the backbone of modern applications. They power everything from e-commerce platforms and financial systems to social media and analytics tools.
Ensuring good database performance is critical, as it directly impacts user experience, operational costs, and the ability to scale when needed.
Users expect applications to respond instantly. A slow database can lead to delays in fetching data, resulting in poor application performance and monetary impact. For example:
Delayed product searches or sluggish checkout processes can frustrate customers, increasing cart abandonment rates in an e-commerce platform.
Slow feed updates can reduce user engagement for a social media application.
A strong database performance is the key to retaining users and keeping them positively engaged with the application.
Database inefficiencies often lead to increased hardware requirements and higher cloud usage bills. Poorly optimized queries can consume excessive CPU, memory, and disk I/O. This leads to organizations provisioning expensive resources to compensate for poor performance, which could have been avoided with proper optimization.
However, optimizing database performance requires a multi-faceted approach. There is no magical one-size-fits-all strategy.
In this post, we’ll explore the various factors that can impact the performance of a database. We’ll also look at multiple strategies that can help improve the database performance.
Key Metrics to Evaluate Database Performance...
Continue reading this post for free in the Substack app
Like
Comment
Restack
© 2024 ByteByteGo
548 Market Street PMB 72296, San Francisco, CA 94104
Unsubscribe
by "ByteByteGo" <bytebytego@substack.com> - 10:35 - 12 Dec 2024 -
What a decade of research reveals about gender diversity at work
Only McKinsey Perspectives
Brought to you by Alex Panas, global leader of industries, & Axel Karlsson, global leader of functional practices and growth platforms
Welcome to the latest edition of Only McKinsey Perspectives. We hope you find our insights useful. Let us know what you think at Alex_Panas@McKinsey.com and Axel_Karlsson@McKinsey.com.
—Alex and Axel
•
A concerning trend. McKinsey’s latest Women in the Workplace report, conducted in partnership with LeanIn.Org, marks a decade of research finding that making progress is difficult—and slow. Women are still a long way off from achieving parity with men; they remain underrepresented both in C-suite roles and across the corporate pipeline, even at the start of their careers. Yet for the first time in ten years, “companies are starting to step back from their commitment” to improve the experience of women at work, McKinsey senior partners Alexis Krivkovich and Lareina Yee reveal on a recent episode of The McKinsey Podcast.
—Edited by Belinda Yu, editor, Atlanta
This email contains information about McKinsey's research, insights, services, or events. By opening our emails or clicking on links, you agree to our use of cookies and web tracking technology. For more information on how we use and protect your information, please review our privacy policy.
You received this email because you subscribed to the Only McKinsey Perspectives newsletter, formerly known as Only McKinsey.
Copyright © 2024 | McKinsey & Company, 3 World Trade Center, 175 Greenwich Street, New York, NY 10007
by "Only McKinsey Perspectives" <publishing@email.mckinsey.com> - 12:10 - 12 Dec 2024 -
🎉 Slack data residency is now available for Singapore
Hi Md Abul,
Salesforce and Slack are committed to meeting your data, compliance, and security requirements. Which is why we are excited to announce that Slack data residency is now available for Singapore. This is an important development for many regulated industries and public sector organisations, and is relevant for any organisation that wants to keep their data within the country.
Slack is where work happens for millions, every day. It’s the conversational interface for Salesforce that will help you to:- Empower your people with agent-powered productivity in one work OS
- Take action with connected CRM data where your teams work
- Drive growth for every line of business from Slack
🚀 Ready to get started with Slack? Contact our Sales team to learn more, and know that your data can be stored in Singapore if that’s a requirement of your organisation. set up data residency for your organisation. To learn more about security at Slack, including data residency, visit Security at Slack page.
Contact sales Kind regards,
Slack from Salesforce© 2024 Salesforce, Inc.
Salesforce.com Singapore Pte Ltd. 5 Temasek Boulevard #13-01 Suntec Tower 5 Singapore 038985
General Inquiries: +65 6302 5700 | Fax: +65 6302 5777 | Sales: 800 1301 448 (Singapore), 800 967 655 (Hong Kong), +65 6302 5700 (Intl)
This email was sent to info@learn.odoo.com
Manage Preferences to Unsubscribe | View as webpage | Privacy Statement
Powered by Salesforce Marketing Cloud
by "Salesforce" <salesforce@mail.salesforce.com> - 08:01 - 11 Dec 2024 -
A gift for you on National App Day
Share this email
Download the app
Scan the QR—or learn more here 
This email contains information about McKinsey’s research, insights, services, or events. By opening our emails or clicking on links, you agree to our use of cookies and web tracking technology. For more information on how we use and protect your information, please review our privacy policy.
You received this email because you are a registered member of mckinsey.com.
Copyright © 2024 | McKinsey & Company, 3 World Trade Center, 175 Greenwich Street, New York, NY 10007
by "McKinsey & Company" <publishing@email.mckinsey.com> - 03:36 - 11 Dec 2024 -
*Quote Request* Bids #
Brilliantly,We specialize in estimating all types of construction projects, providing comprehensive labor and material costs in Excel format. By leveraging our expertise, you can save time and gain a competitive edge, increasing your chances of winning contracts.
Send me the plans for a quote if you have any job for us. You may ask for a sample.
Thanks.
by "San Jose" <sanjoseglobalestimations@gmail.com> - 09:28 - 11 Dec 2024 -
Webcast: What’s ahead for AI-powered ERP?
SAP WEBCAST
Top ERP Trends and Opportunities for 2025 and BeyondHello,
Technology is evolving faster than ever. Wouldn’t it be great to know what’s ahead for the coming year?
Join us for the first event in our 2025 Cloud ERP Essentials series to hear some top predictions from IDC’s upcoming FutureScape report on intelligent ERP.During the Webcast, we’ll discuss the market trends and opportunities for the year ahead. You’ll learn about the impact of generative AI and gain valuable insights to prepare your business for the future.
Our speakers:- Mickey North Rizza, Group Vice President Enterprise Software, IDC
- Lauren Husum, VP of Product Marketing, SAP S/4HANA Cloud Public Edition
Top ERP Trends and Opportunities for 2025 and Beyond
January 15, 2025 | 11:00 a.m. ET/ 11:00 a.m. CET/ 1:00 p.m. SGT
Best regards,
The GROW with SAP team
(121)
Contact us
See our complete list of local country numbers
SAP (Legal Disclosure | SAP)
This e-mail may contain trade secrets or privileged, undisclosed, or otherwise confidential information. If you have received this e-mail in error, you are hereby notified that any review, copying, or distribution of it is strictly prohibited. Please inform us immediately and destroy the original transmittal. Thank you for your cooperation.
You are receiving this e-mail for one or more of the following reasons: you are an SAP customer, you were an SAP customer, SAP was asked to contact you by one of your colleagues, you expressed interest in one or more of our products or services, or you participated in or expressed interest to participate in a webinar, seminar, or event. SAP Privacy Statement
This email was sent to info@learn.odoo.com on behalf of the SAP Group with which you have a business relationship. If you would like to have more information about your Data Controller(s) please click here to contact webmaster@sap.com.
This offer is extended to you under the condition that your acceptance does not violate any applicable laws or policies within your organization. If you are unsure of whether your acceptance may violate any such laws or policies, we strongly encourage you to seek advice from your ethics or compliance official. For organizations that are unable to accept all or a portion of this complimentary offer and would like to pay for their own expenses, upon request, SAP will provide a reasonable market value and an invoice or other suitable payment process.
This e-mail was sent to info@learn.odoo.com by SAP and provides information on SAP’s products and services that may be of interest to you. If you received this e-mail in error, or if you no longer wish to receive communications from the SAP Group of companies, you can unsubscribe here.
To ensure you continue to receive SAP related information properly, please add sap@mailsap.com to your address book or safe senders list.
by "SAP" <sap@mailsap.com> - 07:14 - 11 Dec 2024 -
How can companies become more data driven?
Only McKinsey Perspectives
Brought to you by Alex Panas, global leader of industries, & Axel Karlsson, global leader of functional practices and growth platforms
Welcome to the latest edition of Only McKinsey Perspectives. We hope you find our insights useful. Let us know what you think at Alex_Panas@McKinsey.com and Axel_Karlsson@McKinsey.com.
—Alex and Axel
•
Difficulties with data. The proliferation of generative AI (gen AI) and other technologies offers great promise for organizations, but only if they can get a handle on their data. That isn’t an easy task for many companies. According to McKinsey senior partners Holger Harreis and Kayvaun Rowshankish and their coauthors, 72% of leading organizations report that managing data is one of the top challenges to scaling their AI applications.
•
New kinds of work. To start making the most of their data, companies can expand existing roles within the organization (such as data engineers and data scientists) as well as create new ones. Such roles might include AI ethics stewards, who ensure that AI systems are developed and used responsibly, and specialists who work with unstructured data. Explore the McKinsey Quarterly Five Fifty briefing “The data-driven enterprise” to discover data’s growing business opportunity.
—Edited by Belinda Yu, editor, Atlanta
This email contains information about McKinsey's research, insights, services, or events. By opening our emails or clicking on links, you agree to our use of cookies and web tracking technology. For more information on how we use and protect your information, please review our privacy policy.
You received this email because you subscribed to the Only McKinsey Perspectives newsletter, formerly known as Only McKinsey.
Copyright © 2024 | McKinsey & Company, 3 World Trade Center, 175 Greenwich Street, New York, NY 10007
by "Only McKinsey Perspectives" <publishing@email.mckinsey.com> - 12:24 - 11 Dec 2024 -
How Tinder Recommends To 75 Million Users with Geosharding
How Tinder Recommends To 75 Million Users with Geosharding
Forwarded this email? Subscribe here for moreBuilding AI Apps on Postgres? Start with pgai (Sponsored)
pgai is a PostgreSQL extension that brings more AI workflows to PostgreSQL, like embedding creation and model completion. pgai empowers developers with AI superpowers, making it easier to build search and retrieval-augmented generation (RAG) applications. Automates embedding creation with pgai Vectorizer, keeping your embeddings up to date as your data changes—no manual syncing required. Available free on GitHub or fully managed in Timescale Cloud.
Disclaimer: The details in this post have been derived from the Tinder Technical Blog. All credit for the technical details goes to the Tinder engineering team. The links to the original articles are present in the references section at the end of the post. We’ve attempted to analyze the details and provide our input about them. If you find any inaccuracies or omissions, please leave a comment, and we will do our best to fix them.
Tinder is a dating app that handles billions of swipes daily, matching over 75 million users worldwide.
Their recommendation engine must deliver matches with speed and precision. However, managing a global user base and ensuring seamless performance isn’t easy, especially when searches involve massive amounts of data spread across different regions.
To overcome the challenge, Tinder adopted Geosharding: a method of dividing user data into geographically bound "shards."
This approach enabled the recommendation engine to focus searches only on relevant data, dramatically improving performance and scalability. The system now handles 20 times more computations than before while maintaining low latency and delivering matches faster than ever.
In this post, we’ll explore how Geosharding works, the architecture behind it, and the techniques Tinder uses to ensure data consistency.
The Initial Single-Index Approach
When Tinder started using Elasticsearch to manage its recommendation system, it stored all user data in a single "index".
Think of this index as one massive database holding information about every Tinder user worldwide. While this worked fine when the platform was smaller, it caused significant problems as Tinder grew.
Some major problems were as follows:
The Index Became Too Big: As millions of users joined Tinder, the index grew into an enormous collection of data. Every time someone used the app to find potential matches, the system had to sift through this gigantic index, even though most of the data wasn’t relevant to that search. For example, a user in Los Angeles didn’t have to see people from London, but the system still had to deal with all the unnecessary data, making the process slow and inefficient.
High Latency (Search Slowness): Latency refers to the delay or time it takes to complete a task. In this case, the task was to find appropriate matches. Since the index was so large, it took a lot of time and computing power to search through everything. This delay degraded the user experience.
Rising Costs: To handle the ever-growing workload, Tinder had to add more powerful servers and replicas (backup systems) to maintain speed and reliability. These additions drove up infrastructure costs, making the system expensive to maintain.
The MOST Hands-On Training on AI Tools you’ll ever attend, for free (Sponsored)
The biggest MYTH about AI is that it is for people in tech. Or people who can code.
In reality, AI can be game-changing for you whether you are a curious 11 year old kid wanting to research or a 50 year old professional who is willing to stay relevant in 2024.
Join the 3 hour AI Tools Training (usually $399) but free for the first 100 readers.
By the way, here’s sneak peek into what’s inside the workshop:
Making money using AI 💰
The latest AI developments, like GPT o1 🤖
Creating an AI clone of yourself, that functions exactly like YOU 🫵
10 BRAND new AI tools to automate your work & cut work time by 50% ⏱️
And a lot more that you’re not ready for, just 3 hours! 🤯
1.5 Million people are already RAVING about this hands-on Training on AI Tools. Don’t take our word for it? Attend for yourself and see.
(first 100 people get it for free + $500 bonus) 🎁
The Geosharding Solution
Geosharding was the method Tinder adopted to address the inefficiencies of using a single large index for its recommendation system.
The idea was to divide the global user base into smaller, localized groups, or "shards," based on location. Each shard contains data for users in a specific region, allowing the system to focus its searches only on the most relevant data for a particular query.
How Geosharding Works?
Imagine Tinder’s user base as a map of the world.
Instead of putting all user data into one massive database (like a single, global shelf), the map is divided into smaller sections or shards. Each shard corresponds to a geographic region.
For example:
One shard might include all users within a 50-mile radius of Los Angeles.
Another shard might include users in London.
By organizing users this way, the system can search only the shard that matches a user’s location, avoiding the need to sift through irrelevant data. And all of this was made possible through Geosharding.
Geosharding improves query performance in the following ways:
Focus on Relevant Data: When a user searches for matches, Tinder only interacts with the shards containing users within their specified distance range (e.g., 100 miles). This drastically reduces the amount of data the system needs to process for each query, making searches much faster and more efficient.
Smaller Shard Sizes: Breaking the index into smaller shards results in lighter, more manageable datasets. Smaller datasets mean the system can process queries faster and with less strain on infrastructure.
Better Resource Utilization: By narrowing the search area to a localized shard, computing power is used more effectively. Resources aren’t wasted on processing irrelevant data.
Optimizing Shard Size
Despite the obvious benefits of sharding, finding the "right" size for a shard was crucial for maintaining system efficiency.
Several factors were considered in determining this balance:
Query Efficiency: Shards must be large enough to contain sufficient user data for meaningful matches. Too small a shard size could require the system to query multiple shards, increasing latency.
Shard Migration Overhead: Users don’t stay in one place—they travel, commute, or use features like "Passport" to swipe in different locations. When a user moves to a new geographical area, their data must be migrated to the corresponding shard. If shards are too small, migrations happen more frequently, creating temporary inconsistencies and operational complexity.
User Density Variations: Some areas, like New York City, have high concentrations of users in a small space, while others, like rural areas, have lower densities. Shards need to account for these differences. In densely populated areas, shards may represent smaller geographic regions, whereas in sparsely populated areas, shards might cover larger regions.
Algorithm and Tools Used For Geosharding
To implement Geosharding effectively, Tinder needed tools and algorithms that could efficiently divide the world into geographic shards while ensuring these shards were balanced in terms of user activity.
Two key components made this possible: Google’s S2 Library and a container-based load balancing method. Let’s look at them both in more detail.
The S2 Library: Mapping the Globe Into Cells
The S2 Library is a powerful tool developed by Google for spatial mapping. It divides the Earth’s surface into a hierarchical system of cells, which are smaller regions used for geographic calculations.
Source: S2 Geometry Here’s how it works and why it’s ideal for Geosharding:
Quadtree Structure: The S2 Library organizes cells in a "quadtree" structure, where each cell can be divided into four smaller cells. This allows for a flexible representation of geographic areas, from very large regions to precise, tiny locations.
Spatial Locality with Hilbert Curves: It uses a mathematical concept called a Hilbert curve, which ensures that geographically close points on the Earth’s surface remain close in the cell hierarchy. For example, users in Los Angeles and San Diego, being physically near each other, are also close in the S2 hierarchy. This helps group them logically within the same shard.
The main advantages of S2 for Tinder’s use case were as follows:
Consistent Cell Sizes: S2 ensured that cells at the same level in the hierarchy cover roughly equal areas, avoiding distortions near the poles that other systems (like Geohashes) often face.
Multi-level Granularity: Tinder could choose different cell sizes depending on user density. For instance:
Level 7 cells (~45 miles wide) were ideal for sparsely populated areas.
Level 8 cells (~22.5 miles wide) worked better for dense regions like New York City.
Efficient Geo-queries: Built-in S2 functions allow for quick mapping of a location (latitude/longitude) to a cell or a group of cells that cover a specific area.
Load Balancing: Evenly Distributing Users Across Geoshards
Once the world was divided into S2 cells, the next challenge was to balance the "load" across these cells.
Load refers to the activity or number of users within each shard. Without proper balancing, some shards could become "hot" (overloaded with users), while others remained underutilized.
To solve this, Tinder used a container-based load-balancing approach which involved the following aspects:
Quantifying the Load: Each S2 cell was assigned a load score based on factors like active user count and the number of queries originating from that cell. Higher activity meant a higher load score.
Balancing the Load: Imagine each S2 cell represents a neighborhood, and the “load” score” reflects the number of active Tinder users in that area. For example, a cell with a load score of 10 represents a neighborhood of 10 active users. Tinder used a "container" (a conceptual district) to group nearby neighborhoods into a district. When the container filled up to a certain threshold, it was treated as one Geoshard. This approach ensured that shards were formed based on activity, keeping their total load roughly equal.
Preserving Geographic Locality: Because S2 cells and Hilbert curves maintain spatial locality, the process naturally grouped adjacent cells into the same shard, ensuring each shard represented a contiguous geographic area.
The Abstraction Layer
After finalizing the Geosharding algorithm, Tinder designed a scalable and efficient architecture that relies on an abstraction layer to handle user data seamlessly across Geoshards.
This layer simplifies interactions between the application, the recommendation system, and the geosharded Elasticsearch cluster.
See the diagram below:
Here’s how the setup works:
Abstraction Layer: The abstraction layer is an internal microservice that acts as a bridge between Tinder’s application logic and the geosharded Elasticsearch system. It hides the complexity of Geosharding from the application layer, allowing the recommendation engine to interact with the cluster without worrying about the underlying shard configurations or user data migration.
Location Updates and Shard Movement: When a user’s location changes (for example, when traveling to a new city), the abstraction layer determines if this change requires the user’s data to be moved to a different Geoshard. It accesses Tinder’s location services to evaluate the user’s new position and initiates a shard move if necessary. The abstraction layer ensures the data is removed from the old shard and correctly added to the new one.
Query Handling: When a user searches for matches, the recommendation engine sends a request to the abstraction layer. The abstraction layer calculates which Geoshards need to be queried based on the user’s current location and their distance filter (e.g., a 50-mile radius). It distributes the query to the appropriate shards in the Elasticsearch cluster, collects the results, and aggregates them before sending them back to the client.
Geosharded Indices: The actual data is stored in Geosharded indices within the Elasticsearch cluster. These indices represent logical shards, divided based on geographic regions, and are accessed through the abstraction layer. The abstraction layer maintains a mapping of users to their respective Geoshards, ensuring that all operations (like searches or data migrations) are directed to the correct shard.
Dynamic Adaptation: The abstraction layer simplifies operations like shard rebalancing and re-sharding. For example, if user density in a specific area grows or shrinks significantly, the abstraction layer helps reconfigure shards to accommodate the changes without disrupting the system.
Multi-Index vs Multi-Cluster
When building the Geosharded recommendation system, Tinder had to decide how to organize its data infrastructure to manage the geographically separated shards efficiently.
This led to a choice between two approaches: multi-index and multi-cluster.
Each approach had pros and cons, but Tinder ultimately chose the multi-index approach. Let’s break it down in simple terms.
What is the Multi-Index Approach?
In the multi-index setup:
All the shards (Geoshards) are stored as separate indices within a single Elasticsearch cluster.
Each index corresponds to a logical Geoshard, containing user data for a specific geographic area.
The alternative was a multi-cluster setup that involved creating separate Elasticsearch clusters for each geoshard. While this approach has some advantages, it wasn’t the right fit for Tinder due to a lack of native support for cross-cluster queries and higher maintenance overhead.
To overcome the limitations of the multi-index setup, particularly the risk of uneven load distribution, Tinder implemented appropriate load balancing techniques. Some of the details about these techniques are as follows:
Dynamic Scaling Through Replica Adjustments:
A replica is a copy of an index shard. Adding more replicas distributes the workload across multiple servers, reducing the burden on any single shard.
Tinder adjusted the number of replicas dynamically based on the activity levels in each Geoshard:
High-traffic Geoshards (e.g., urban areas with many users) were given more replicas to handle the extra load.
Low-traffic Geoshards (e.g., rural areas) required fewer replicas, saving resources.
Handling Load Imbalances:
If a particular Geoshard became a "hotspot" due to a surge in user activity, its replicas could be increased temporarily to balance the load.
This ensured consistent performance across all regions, even during unexpected spikes in usage.
Fault Tolerance and Scalability:
By spreading replicas across multiple nodes in the cluster, Tinder ensured that the system could handle server failures without interrupting service.
This setup also allowed Tinder to scale horizontally by adding more servers to the cluster as needed.
Handling Time Zones: Balancing Traffic Across Geoshards
One of the key challenges Tinder faced with Geosharding was the variation in traffic patterns across Geoshards due to time zones.
See the diagram below that shows the traffic pattern of two Geoshards during a 24-hour time span:
Source: Tinder Tech Blog Users within the same geoshard are typically in the same or adjacent time zones, meaning their active hours tend to overlap. For example:
A shard covering New York experiences peak traffic during the evening in the Eastern Time Zone.
A shard covering Tokyo might see peak traffic several hours later, during the evening in Japan.
If shards were assigned directly to physical servers without considering time zone effects, some servers would be overloaded during peak hours for one shard, while others would remain idle.
To solve this problem, Tinder implemented a randomized distribution of shards and replicas across physical nodes in the Elasticsearch cluster.
Here’s how it works and why it helps:
Random Allocation of Shards: Each physical server (or "node") hosts multiple Geoshards. These shards are randomly distributed across the available nodes. For example, a single server might hold:
A primary shard for New York.
A replica shard for Tokyo.
Another replica shard for London.
Load Balancing Through Time-Zone Offsets: Since Geoshards covering different time zones have their peaks at different times of the day, random distribution ensures that the load on each node is more evenly spread throughout a 24-hour period. A server hosting New York’s shard (with peak traffic in the evening) will also host Tokyo’s shard (with peak traffic at a different time), balancing the load spikes.
Resilience to Changes: The random distribution of shards also helps in case of unexpected traffic surges. If one Geoshard suddenly experiences higher activity, the other shards on the same node are less likely to be at their peaks, preventing server overload.
The Overall Cluster Design
Tinder’s Geosharded recommendation system was built using a carefully designed cluster architecture to handle billions of daily swipes while maintaining speed, reliability, and scalability.
The cluster architecture divides the responsibilities among different types of nodes, each playing a specific role in ensuring the system’s performance and fault tolerance.
See the diagram below:
The key components of the cluster are as follows:
Master Nodes
These are the managers of the cluster, responsible for overall health and coordination.
They keep track of which shards are stored on which data nodes. They also monitor the status of the nodes and redistribute shards if a node fails.
Master nodes don’t handle user queries directly. Their role is to manage the system and ensure everything runs smoothly.
Coordinating Nodes
The coordinating nodes act as the traffic controllers of the system.
When a user performs a search or swipe, their request goes to a coordinating node. This node determines:
Which Geoshards need to be queried based on the user’s location and filters?
Which data nodes hold the relevant shards to process the request?
Once the results are collected from the data nodes, the coordinating node aggregates them and sends the final response back to the user.
Data Nodes
These are the workers of the cluster, responsible for storing the actual data and processing queries. Data nodes execute the search queries sent by coordinating nodes and return the results.
Each data node holds multiple Geoshards and their replicas as follows:
Primary Shards: The main copies of data.
Replica Shards: Backup copies that improve fault tolerance and help balance the workload.
Consistency Challenges
Maintaining data consistency was another significant challenge in Tinder’s Geosharded system.
This complexity arises because users frequently move between locations, and their data must be dynamically shifted between Geoshards. Without proper handling, these transitions could lead to inconsistencies, such as failed writes, outdated information, or mismatches between data locations.
The key challenges are as follows:
Concurrent Updates: In Tinder’s real-time environment, multiple updates to the same user data can happen within milliseconds of each other. For example, a user could use Tinder’s Passport feature to switch their location to another city and then quickly return to their original location. If updates are not processed in the correct order, the system could end up storing outdated or incorrect information.
Failed Writes: During the migration of user data between Geoshards, network issues or system overloads could cause write operations to fail, leaving the data incomplete or missing.
Stale Data: A delay in updating Geoshard mappings could result in queries pointing to the wrong shard, leading to the retrieval of outdated information.
Mapping Mismatches: The mapping datastore, which keeps track of which shard a user belongs to, could fall out of sync with the actual location of the user’s data, causing future updates or queries to fail.
Tinder implemented a combination of strategies to address these issues and ensure consistent, reliable data handling across Geoshards. Let’s look at a few of those techniques in detail.
1 - Guaranteed Write Ordering with Apache Kafka
When multiple updates occur for the same user data, ensuring they are processed in the correct order is critical to avoid inconsistencies.
Apache Kafka was used as the backbone for managing data updates because it guarantees that messages within a partition are delivered in the same order they were sent.
Tinder assigns each user a unique key (for example, their user ID) and uses consistent hashing to map updates for that user to a specific Kafka partition. This means all updates for a particular user are sent to the same partition, ensuring they are processed in the order they are produced.
See the diagram below:
The consumers (parts of the system that read and process messages) read updates sequentially from the partitions, ensuring that no updates are skipped or processed out of order.
This guarantees that the latest update always reflects the most recent user activity.
2 - Strongly Consistent Reads with the Elasticsearch Get API
Elasticsearch is a "near real-time" search engine, meaning recently written data may not be immediately available for queries. This delay can lead to inconsistencies when trying to retrieve or update data.
Tinder, however, leveraged Elasticsearch’s Get API, which forces the index to refresh before retrieving data. This ensures that any pending updates are applied before the data is accessed.
Reindexing and Refeeding:
In cases where data inconsistencies occurred due to upstream failures or mismatches between shards, a mechanism was needed to realign the datastore with the source of truth.
Data from the source shard was copied to the target shard when a user moved between Geoshards.
If inconsistencies were detected (e.g., missing data in a shard), the system periodically re-synced the search datastore with the source datastore using a background process.
This ensured that even if temporary errors occurred, the system would self-correct over time.
Conclusion
Tinder’s implementation of Geosharding demonstrates the complexities of scalability, performance, and data consistency in a global application.
By dividing its user base into geographically bound shards, Tinder optimized its recommendation engine to handle billions of daily swipes while maintaining lightning-fast response times. Leveraging tools like the S2 Library and Apache Kafka, along with algorithms for load balancing and consistency, the platform transformed its infrastructure to support a seamless user experience across the globe.
This architecture improved performance by handling 20 times more computations than the previous system. It also addressed challenges like traffic imbalances across time zones and potential data inconsistencies during shard migrations. With randomized shard distribution, dynamic replica adjustments, and intelligent reindexing, Tinder ensured reliability, fault tolerance, and scalability.
References:
SPONSOR US
Get your product in front of more than 1,000,000 tech professionals.
Our newsletter puts your products and services directly in front of an audience that matters - hundreds of thousands of engineering leaders and senior engineers - who have influence over significant tech decisions and big purchases.
Space Fills Up Fast - Reserve Today
Ad spots typically sell out about 4 weeks in advance. To ensure your ad reaches this influential audience, reserve your space now by emailing sponsorship@bytebytego.com.
© 2024 ByteByteGo
548 Market Street PMB 72296, San Francisco, CA 94104
Unsubscribe
by "ByteByteGo" <bytebytego@substack.com> - 10:35 - 10 Dec 2024 -
From China to Arizona: The oddities of global clocks
From China to Arizona: The oddities of global clocks
Most common global data challenges in 2024
Monthly changes
🕛 The world’s most confusing clocks 🕛
CUSTOMER STORIES
The 4 Most Common Global Location
Data Challenges in 2024
Let's close the year by compiling our customer's most common location data challenges and how they solved them.
How to validate postal data 300X faster
SUPPLY CHAIN AND LOGISTICS
Manually finding and validating global postal data takes time and effort. Larger cities can have hundreds of zip codes, and some countries change them regularly. Errors can lead to shipping delays and delivery problems down the line.
Improving targeting with geocoded postal data
GEOMARKETING
Inaccurate geofencing will affect the performance of your marketing activities and ad spending, impacting your revenue stream.
Address Matching Using Reliable Postal Data
MARKETPLACES
Maintaining a global in-house address database is challenging both time-wise and cost-wise. This often leads to dirty addresses and gaps in the data, hindering the customer experience of your website.
Mapping risk areas to postal boundaries and points
INSURANCE
Inaccurate and outdated zip code boundaries can lead to failed identification of addresses at risk, leaving insurers open to unnecessary losses.
MONTHLY CHANGES
Canada postal codes are notoriously difficult to localize. Thanks to a new public data source, the coordinates of thousands of points have been improved.
Full Postal & Street updates
Canada, Israel, Netherlands, Norway, Portugal, Puerto Rico, Romania, United Kingdom
Postal database patches
(small changes)
Brazil (Coordinates)
Postal boundaries updates
United States, South Africa
Postal boundaries patches (small changes)
Armenia, Belgium, Brunei, Brazil, Guatemala, India, Ireland, Netherlands, San Marino, Thailand, United Kingdom, Vatican
Administrative boundaries patches (small changes)
Argentina
For a detailed changelog, take a look at the CSV file.
The world's most confusing clocks
Time zones reveal the fascinating intersection of geography and culture!
Despite its size, China uses one time zone. This means that in Tibet, the sun doesn't rise until 10 a.m. in winter, while Afghanistan is 3.5 hours behind.
Arizona has its own time mess, with some areas observing daylight savings and others not—so your watch may flip as you cross state lines.
The Diomede Islands, only 2.5 miles apart in the Pacific, are a full day apart due to the International Date Line.
Kiribati jumped across the Date Line in 1995, skipping December 31 altogether to sync with Australia's business hours.
Time zones aren't just about clocks—they're a reflection of human culture, borders, and, occasionally, chaos. ⏳
Source: Map Men
Follow us on LinkedIn for more geographical facts like this!
Kind regards,
Jérôme & the GeoPostcodes team
PS: Interested in previous Monthly Product Updates? Read here.
GeoPostcodes, Bld Bischoffsheim 15, 1000 Bruxelles, Belgium
by "Jérôme from GeoPostcodes" <jerome@geopostcodes.com> - 05:46 - 10 Dec 2024 -
RE : Seeking for Argentina agent
Dear agent,
Greeting from Cara and GLA family. Trust you are doing great.
I sincerely invite you to join GLA platform and cover the increasing request from/to Argentina from our GLA members.
Above 75% of GLA Members are first-class agents with wide range of primary resources. GLA fulfills all your needs for your business development and resources integration, in terms of finding reliable agents and getting huge volume of inquiries from/to your area.
Besides, We are planning to hold the 12th GLA Global Logistics Conference:
Ø Event Name: The 12th GLA Global Logistics Conference
Ø Host By: GLA Global Logistics Alliance
Ø Event Date: May 15 - 18, 2025 (4 days)
Ø Host City: Dubai, UAE
Ø Expected Attendee: 1500+
Ø From: 130+ countries
Ø 1-on-1 Meeting: 51 times per pax
The early bird price for the 12th GLA Conference is now open till 31st Dec 2024! Please don’t miss this opportunity to attend the conference as a GLA member.
If you want to know more, please don’t hesitate to contact me.
Best regards,
Cara Chen
GLA Overseas department
Mobile:
(86) 190-7616-6926 (WhatsApp|Wechat)
Email:
Company:
GLA Co.,Ltd
Website:
Address:
No. 2109, 21st Floor, HongChang, Plaza, No. 2001, Road Shenzhen, China
The 12th GLA Conference – Dubai, UAE on 15th - 18th May 2025 – click here for registration
Ø The 11th GLA Conference in Bangkok Thailand, online album
Ø The 10th GLA Conference in Dubai UAE, online album
Ø The 9th GLA Conference in Hainan China, online album
【Notice Agreement No 7】
7. GLA president reserves the right to cancel or reject membership or application.
company shall cease to be a member of GLA if:
a) the Member does not adhere to GLA terms and conditions
b) the Member gives notice of resignation in writing to the GLA.
c) No good reputation in the market.
d) Have bad debt records in the GLA platform or in the market
by "Cara" <member226@glafamily.com> - 05:15 - 10 Dec 2024 -
Hi there, can I send you some examples?
Hi there, can I send you some examples?
Hi there,
At CRM Management Ltd, we're passionate about designing innovative CRM systems that perfectly fit the unique needs of our clients. Our track record speaks volumes, having successfully collaborated with businesses of all sizes across various industries.
Can I send you some examples of some recent systems we've produced?
Here are some of the reasons our clients choose us, and stick with us:
- Understand your business: Our solutions are built based on real-life usage and feedback, ensuring they align with your unique workflows.
- Building partnerships: Beyond delivering projects, we cultivate long-term partnerships with our clients, supported by our robust project management practices.
- Flexibility: Our agile approach allows us to adapt quickly to meet your evolving needs, delivering solutions promptly and effectively.
- Clear pricing: Our transparent pricing structure ensures there are no hidden fees, with support packages tailored to your budget and requirements.
Whether you're new to CRM systems or contemplating an upgrade, it would be great to have a chat to understand your requirements.
Let me know if you would like to see some examples.
Kind regards,
Andrew Miller
CRM Management Ltd
01759 361 845
crmmanagement.co.uk
I have sent this email to info@learn.odoo.com. If you don't want to get any more emails from us, you can unsubscribe here.
CRM Management Ltd - English House, Railway St, Pocklington, York YO42 2QZ - Company no. 10684534
by "Andrew Miller" <Andrew@crmcompany.co.uk> - 05:10 - 10 Dec 2024 -
The technological leaps that have changed business forever
Only McKinsey Perspectives
Brought to you by Alex Panas, global leader of industries, & Axel Karlsson, global leader of functional practices and growth platforms
Welcome to the latest edition of Only McKinsey Perspectives. We hope you find our insights useful. Let us know what you think at Alex_Panas@McKinsey.com and Axel_Karlsson@McKinsey.com.
—Alex and Axel
•
Turning points in business tech. Much has changed since the McKinsey Quarterly was founded in 1964. In honor of our flagship publication’s 60th birthday, we revisited more than six decades of history to identify 25 turning points in technology that have transformed the face of business—from the semiconductor revolution of the 1960s to the birth of the World Wide Web in the 1990s and the current decade’s advances in space exploration and AI.
•
Leaps and bounds. While we’re only a few years into the 2020s, it’s already been a banner decade for technological advances. In 2021, commercial space exploration took a giant leap forward: an all-civilian crew went into outer space’s orbit for the first time ever. The next year, 2022, witnessed a bold new development in technology with OpenAI’s release of ChatGPT, its generative-AI-based chatbot that’s capable of holding humanlike conversations. Visit the interactive for a visual history of the key moments in business technology, and sign up for a free digital Quarterly membership to access our commemorative issues.
—Edited by Belinda Yu, editor, Atlanta
This email contains information about McKinsey's research, insights, services, or events. By opening our emails or clicking on links, you agree to our use of cookies and web tracking technology. For more information on how we use and protect your information, please review our privacy policy.
You received this email because you subscribed to the Only McKinsey Perspectives newsletter, formerly known as Only McKinsey.
Copyright © 2024 | McKinsey & Company, 3 World Trade Center, 175 Greenwich Street, New York, NY 10007
by "Only McKinsey Perspectives" <publishing@email.mckinsey.com> - 12:12 - 10 Dec 2024 -
Re: Guest Post Contribution Inquiry
Re: Guest Post Contribution Inquiry
Hi,
I hope this message finds you well. I'm just reaching out to quickly follow up on our previous correspondence. Please feel free to ignore this email if the matter doesn't interest you.
Thank you for your time and consideration.
--
On Wed, Dec 4, 2024 at 5:51 PM "Ricardo Gonzalez" <ricardo.g@zupo.co> wrote:Hi,
I’d like to pitch David Maricich, President and Chief Creative Officer of Maricich Health, as a contributor for Odoo. David is a visionary leader and healthcare entrepreneur who has dedicated over 20 years to consulting on healthcare innovation, branding, marketing, and communications.
David can provide valuable insights into effective healthcare marketing during major transitions, such as mergers and acquisitions, and can share strategies for reaching diverse target markets and enhancing public health campaigns. Additionally, he is keen to discuss the transformative potential of artificial intelligence in healthcare marketing and innovative strategies for unlocking growth potential in healthcare organizations.
Here are possible topics we’d like to contribute:
- Future-Proofing a Legacy: The Role of Innovation in Multi-Generational Companies
- How to Bring Innovation & Change Without Losing Legacy Values in Your Business
- Adapting to Cultural Fluidity: Strategies for Connecting with Evolving Consumer Identities
- How Marketing Leaders Can Drive Transformation in Stubborn & Traditional Enterprises
- Less is More: How Focused, Specialized Positioning Can Unlock Business Growth
Here are some of his writing samples:
- Could Marketing in Schools Help Save America’s Health? - MM&M Online
- Why the Healthcare Ecosystem Matters - DTC Perspectives
Let me know if any of these topics are of interest to Odoo!
P.S. I coordinate the content for Maricich Health and our contributor posts. David will be the one writing the piece, and I will coordinate edits.
--
by "Ricardo Gonzalez" <ricardo.g@zupo.co> - 11:30 - 9 Dec 2024
