Functionality: The hosted search service allows you to do the following:
- Create and configure an index, and upload your data;
- Configure searchable attributes;
- Enable instant search (search as you type);
- Allow error-correction on the fly;
- Achieve a high query throughput;
- Support real-time updates.
Motivation and Technology: Our motivation is based on the observation that emerging applications have new search requirements that cannot be met by existing solutions such as Lucene and Sphinx. Specifically:
- More users are accessing information from mobile devices, which have the “fat fingers problem,” i.e, tapping is hard and error prone.
- Fuzzy search is important when users do not remember the exact spelling of keywords such as a name or a restaurant.
- Data in many domains (especially social data) is very dynamic, and requires real-time indexing.
- The increasing popularity of Google Instant shows the power of instant search. At the same time, there were no good solutions for developers to do this type of search on large amounts of data with high efficiency and good ranking.
In order to meet these requirements, we develop a hosted service using our own search engine and technology, which is a result of many years of both academic research and commercial development. We use state-of-the-art in-memory index structures and search algorithms specifically designed to support these powerful features.
Comparisons with Existing Solutions: Compared to open source solutions such as Lucene, our engine supports instant, fuzzy search on large amounts of data with a high efficiency to ensure high-quality results. The engine also allows real-time indexes for dynamic data. A detailed comparison with Lucene is available in this white paper.
We also did a comparison with Amazon Cloud Search. For both services, we uploaded 2 million records, and each record has 5 keywords on the average. We used Jmeter to test the performance of both services of instant queries (i.e., prefix queries). Here are the results.
|Index construction time||Search throughput||Average response time||Time to update 200,000 records||Error correction|
|Amazon Cloud Search||47 minutes||752/minute||74 ms||720 seconds||Not supported|
|Bimaple||3 minutes||762/minute||73 ms||21 seconds||Supported|
Our initial comparison shows several advantages of our hosted search: (1) Significantly lower index construction time; (2) Much faster incremental updates of the indexes; and (3) Efficient support of error correction (versus no error correction at all), all without compromising search throughput and response time.
Pricing: Our service is initially free for customers with less than 100MB data for an index. If your data is more than this limit, please contact us.
For more information, please email firstname.lastname@example.org.