Olaworks, Inc. is a Korean computer vision company that creates face and object recognition technology. Olaworks' products include camera-based applications for mobile devices and a visual search platform called ScanSearch. In addition to recognition, the company's technologies perform detailed image and video annotation, such as indexing and tagging. Olaworks was founded in 2006 and currently holds more than 60 patents.

When Olaworks began developing its architecture, the company estimated that it would need to acquire 100 servers at a cost of approximately $500,000, with associated maintenance and operating fees of $10,000 per month.The company would also need to hire at least three system administrators to help maintain the servers. Instead of making such a large investment in hardware and staff, the company turned to Amazon Web Services (AWS).

Olaworks' large-scale image recognition system can register billions of images and recognize an individual image within one second. The AWS-based infrastructure supporting this impressive system consists of one hundred Amazon Elastic Compute Cloud (Amazon EC2) instances, rather than 100 physical servers. Olaworks' Amazon EC2 instances use the Auto Scaling feature to help manage unpredictable usage increases and decreases.

To house its vast amount of image files and associated data, the company uses Amazon's highly-scalable storage option, Amazon Simple Storage Service (Amazon S3), as well as Amazon Elastic Block Store (Amazon EBS), which provides block level storage volumes for individual Amazon EC2 instances.

Amazon's flexible, non-relational data store, Amazon SimpleDB, manages Olaworks' annotation data. This collection of data is particularly complex because different categories require unique fields of information, such as Title and Author fields for an image of a book, and Actor and Producer fields for a movie clip. Kim Tae-hoon, Olaworks' Chief Scientist, explains, "It is well known that efficiently developing a system for storing and retrieving data with various fields is not easy. However, we achieved this by simply using Amazon SimpleDB, which provides a way to handle this type of data while still guaranteeing a high level of performance."

When Olaworks' engineers develop a new feature, they must test with millions of images before implementation. If the company relied on a small collection of physical servers as a pre-production environment, such testing could take several days. However, the company conducts its testing in Amazon Elastic MapReduce in just a few hours. Amazon Elastic MapReduce's Hadoop framework runs on Amazon EC2 instances and provides instant capacity for such data-intense computing.

Not only does Olaworks credit Amazon Elastic MapReduce with helping to decrease its testing time, the service allows the company to introduce a wider variety of features while improving the core technology. Kim Tae-hoon says, "Because of the Amazon Elastic MapReduce, we can try more features and maximize the performance of our image recognition."

Olaworks still maintains a number of applications in a traditional datacenter. The company's developers spend roughly ten percent of their time maintaining these physical servers. Olaworks believes this is an unnecessary burden and is in the process of moving its entire system to AWS.

Olaworks estimates that AWS will help the company save $650,000 in the near term, with the potential to save millions more in coming years. This savings, paired with the stability of AWS, are key ingredients in the development of exciting new recognition technologies. Kim Tae-hoon says, "Instead of focusing our attention on the reliability of our system, AWS helps us focus on accelerating the improvement of its performance and quality."

To find out more about how AWS can help you store and process big data, visit our Big Data details page: http://aws.amazon.com/big-data/.