EHarmony and MongoDB

Thod Nguyen, CTO of the EHarmony company, showed us a leap-forward in the user experience of the world's largest dating service site. After using the MongoDB technology, the pairing speed of eharmony is increased by 95% compared with the original relational database, and the website is increased by 50% subscriptions.

eharmony's main business now is in North America, Australia and the UK. Since the company was founded in 2000, the company has maintained a good record of service--------more than 1.2 million pairs of successful couples finally entered the palace of marriage. Up to today, eharmony already has 55000000 registered users. In the near future, their business is about to expand to another 20 countries, and it is believed that the figure will soon grow rapidly again.

EHarmony employs some fairly professional data engineers to meet many of the future partners. Users of the site will be asked to fill out a detailed questionnaire at the time of registration, and their answers will be used for intelligent mode analysis and then a personal profile for that user. Subsequent data research is more about machine learning and analysis predictions, which will be applied to the final user-pairing algorithm to improve the matching hit rate.

The process of user pairing is not as simple as Google searching for an item or statement, which is "bidirectional" and involves cross-referencing and scoring many attributes, such as age, location, educational background, income, and so on.

In the original architecture of eharmony, the whole system used only a single monolithic database to store all the user data and pairing information, which obviously did not meet the needs of the development of the website business. Later, they began to use the Postgres database to separate the pairing information, which certainly solved their problem, but as the site's potential pairing data reached 3 billion, 25TB data/days, it would take at least 2 weeks to run a complete pairing analysis of a user base. At the same time, with the increase of data volume, the data model becomes more and more complex. And the adjustment of algorithm strategy needs the whole database data emptying and overloading, which not only increases the operation complexity and downtime, but also hinders the rapid development of the company's business.

eharmony understand that they need a new approach and that their databases need to have:

Support complex multi-attribute queries, which are the fundamental needs of their pairing analysis system.

--a more flexible data model, which allows them to use new attributes more precisely.

--large-scale deployment of common hardware devices without adding to existing device overhead.

eharmony once chose Apache SOLR as a possible solution, but because of two-way search rather than one-way search, the plan to select Apache Solr ran aground. After that, Apache Cassandra also because the API is difficult to compatible with the existing data model, read and write operation of the imbalance and other reasons to quit the competition.

In the end, after careful evaluation, eharmony chose MongoDB.

In addition to meeting the above three requirements, MongoDB also provides support for many enterprise Services eharmony through Mongodbenterprise.

Thod through this MongoDB and eharmony's cooperation, also summed up the reader's several views:
-Early use of MONGODB,MONGODB can provide excellent data model applications, segmentation, and release of products.

-When testing, the use of product data and query function. Randomly kills nodes to test the response of the database to multiple failure situations.

--run in stealth, compare the existing relational database to understand the performance of MongoDB processing large amounts of data.

Of course, MongoDB is not eharmony's entire data-management architecture, and their team of data engineers also combines MongoDB with Hadoop, in addition to the Apache Spark and R languages.

The ROI of this database conversion has been validated:

--95% faster. The pairing speed was increased by 95%, and the speed of matching the entire user base was shortened from 2 weeks to 12 hours.

--30% higher. The potential pair increased their communication by 30%.

--50% increased. Paid users increased by 50%.

--60% increased. A 60% increase in the number of visits to a single website.

Of course, the story doesn't end so soon. As eharmony prepares to expand to the other 20 countries, they also intend to expand their data research teams. At the same time they will begin to add location-based services to enrich their mobile-side experience, which coincides with MONGODB support for geo-spatial indexing and querying.