HBase - A case study GAP

Please feel free to contact us if you have any queries.

A Global Fashion Brand :  GAP Inc

How can the horizontal scalability of HBase help the business grow?

Why are the strong-consistency-model, auto-sharding and auto-failover so important to Live Websites?

Problem domain

Business grows, BEFORE 2005, only 3 independent brands, AFTER 2010, has 5 integrated brands: Rapid expansion of apparel catalog, each brand/market combination requires a logical catalog store.

What are wanted

An integrated catalog store for all brands/markets combination, each brand/market combination requires a separate and logical catalog store, highly available, minimal caching because of the problem to keep caches in sync.

Initial Explorations

Sharded RDBMS, MemCached: significant effort was required, more important, still had scalability limits
HBase POC: promising results - decided to move ahead

Why HBase

Strong Consistency Model
Server Side Filters
Automatic sharding 
Automatic Failover
Hadoop Integrated out of box

Architecture 

HBase cluster: 16 HBase slave servers and 3 HBase master servers

Read Mostly: Website Traffic
Write/Delete Bursts: Catalog publishing
Continuous writes: Inventory updates

Phased out originating system, to near real-time-update HBase platform
Near real time updates of Inventory, pricing, and items, therefore, out of sync caches are eliminated

Full product graph in one read
Single path of graph from root to leaf node
Search with secondary indices

Enabled cross-brand selling (sibling to-and-from sibling)

The horizontal scalability and high availability of HBase enables rapid expansion

Single and integrated Catalog store enables cross brand business features !

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.