Database Reference
In-Depth Information
able us to use Cassandra for multiple mixed read-write workload use cases with
strict read-latency requirements. Also, Cassandra's tunable consistency enables us
to have full control over the availability, partition tolerance, and consistency equa-
tion. This capability has enabled us to design many diverse use cases. Another
less-talked-about benefit is that Cassandra's peer-to-peer architecture gives each
node the same responsibility and roughly equal traffic load. This enables better
utilization of hardware for us because we don't need to have nodes sit idle hold-
ing cold replicas or only serving read traffic. Each node in Cassandra serves reads,
handles writes, and acts as replica.
Cassandra Growth
We started in late 2011 with one Cassandra cluster of eight nodes serving one use
case. Now, our Cassandra deployment is more than ten clusters with more than
100 nodes spanning multiple data centers. We have over 250TB provisioned on
local and shared HDDs as well as SSDs. At the time of this writing, we are getting
over nine billion writes and over five billion reads in production and the volume is
growing nearly exponentially.
Many Use Cases
Cassandra has enabled us to do many things that were not practical earlier. We
have built a near-real-time graph-based recommendation system on top of Cas-
sandra. This helps us compute users' taste profiles in near real time. We use
Cassandra for many time-series-data-based use cases in which processing high-
volume, real-time data is a top priority. These include fraud detection, order and
shipment tracking, insights across many applications, pricing engine for affiliates,
and mobile notification tracking, just to name a few. Another use case involves en-
abling social signals on eBay product pages, including like/own/want buttons and
system originally based on Oracle and MySQL in-memory DBs over to Cassandra.
This personalization system serves real-time personalization to a user based on his
or her activities on
eBay.com
such as browsing, buying, or selling. We also use
Cassandra to store raw and aggregated metrics from thousands of production ma-
chines for monitoring and alerting purposes. Our cloud management system uses
Cassandra to store cloud configuration change history. In addition, our sister com-
panies such as RedLaser and Milo are heavy Cassandra users. They also have a
large list of use cases. Since they can't all be covered here, these are two of the
more interesting ones:
