Scaling with HiveDB

Project Genesis
• Cafepress.com Product Catalog
• Hundreds of Millions of Products
• Millions
week
of new products every

• Accelerating growth
Enter Jeremy and
HiveDB
 Our Requirements
• OLTP Optimized
• Constant response time is more
important than low latency
• Related sets vary wildly in size
• Growth hotspots
• Usage hotspots
Partition by key
 Directory
• No broadcasting
• No re-partitioning
• Easy to relocate records
• Easy to add capacity
 Disadvantages
• Intelligence moved from the
database to the application
(Queries have to be planned and
indexed)
• Can’t join across partitions
• NO OLAP! (We consider this a
benefit.)
• Directory is a bottleneck
 Original Design
• Smallest possible implementation
• HiveDB was just a JDBC Gatekeeper
for applications.
 Development
Complexity Problem
• You have to maintain
synchronization between the
directory and the data nodes.
• Lots of code for simple operations
• Data access objects have to be re-
implemented
 Enter Hibernate
Shards
• Partitioned Hibernate from Google
• Why did we write this thing
again?
• Oh wait, you have to tell it how
to look things up...we’re good at
that.
Benefits of Shards
• Unified data access layer
• Result set aggregation across
partitions
• Everyone in the JAVA-world knows
Hibernate.
Show don’t tell!
Competitive Landscape

• Clustered Relational Databases

• Non-relational
Databases
Structured

• Non-relational,
Storage
Unstructured
 Competitive Landscape
Clustered Non-relational Unstructured

Oracle RAC
MS SQL Server Hypertable
Hadoop
MySQL Cluster HBase
MogileFS
DB2 CouchDB
S3
Teradata (OLAP) SimpleDB
HiveDB
 Competitive Landscape
Storage Interface Partitioning Expansion Node Types Maturity

Oracle RAC Shared SQL Transparent No downtime Identical 7 years

Memory / Requires
MySQL Cluster SQL Transparent Mixed 3 years
Local Restart

?
Hypertable Local HQL Transparent No downtime Mixed (released
2/08)
3 years
Degraded
DB2 Local SQL Fixed Hash Identical (25 years
Performance
total)

Key-based 18 months
HiveDB Local SQL No downtime Mixed
Directory (+13 years!)
 Case Study:
CafePress
• Leader in User-generated Commerce
• Same number of products as eBay
(>150,000,000)
• 24/7/365
 Case Study:
Performance Requirements
• Thousands of queries/sec
• 10 : 1 read/write ratio
• Geographically distributed
 Case Study:
Test Environment
• Real schema
• Production-class hardware
• Partial data (~40M records)
CafePress HiveDB 2007
Performance Test Environment

JMeter (1 thread) JMeter (no threads)

command &

client.jar client.jar
control

Measurement Test Controller

Workstation Workstation
100MBit switch
load generators

JMeter (100s of threads) JMeter (100s of threads)

client.jar client.jar

48GB backplane non-

Dell 2950 / 2x2 Xeon blocking gigabit switch Dell 2950 / 2x2 Xeon
16GB, 6x72GB 15k 16GB, 6x72GB 15k
web service

Hardware LB
(hivedb)

Dell 1950 / 2x2 Xeon Dell 1950 / 2x2 Xeon Dell 1950 / 2x2 Xeon
Tomcat 5.5 Tomcat 5.5 Tomcat 5.5

Directory Partition 0 Partition 1

databases
(mysql)

Dell 2950 / 2x2 Xeon Dell 2950 / 2x2 Xeon Dell 2950 / 2x2 Xeon
16GB, 6x72GB 15k 16GB, 6x72GB 15k 16GB, 6x72GB 15k

[email protected] Modified on April 09 2007

 Case Study:
Performance Goals
• Large object reads: 1500/s
• Large object writes: 200/s
• Response time: 100ms
 Case Study:
Performance Results
• Large object reads: 1500/s
Actual result: 2250/s
• Large object writes: 200/s
Actual result: 300/s
• Response time: 100ms
Actual result: 8ms
 Case Study:
Performance Results
• Max read throughput
Actual result: 4100/s
(CPU limited in Java layer;
MySQL <25% utilized)
 Case Study:
Organizational Results
• Billions of queries served
• Highest DB uptime at CafePress
• Hundreds
performed
of millions of updates
High Availability &
Replication
• We don’t specify a fail over
strategy
• We delegate to MySQL replication
Non-JAVA Deployment
Options
• Web service
• JVM Dynamic Languages
HiveDB Accessories
 Class Generation
• Automatically generate Data
Transfer Objects from interfaces
(and soon web services).
 Blobject
• Gets around the problem of ALTER
statements
• Compression
• The hive can contain multiple
versions of a serialized record.
• No data set of this size can be
transformed synchronously.
 Features Teaser
• We’re taking over HA...you’re still
on your own for replication.
• Generated Web Services
• Monitoring
graphs!)
& RRD stats (with

• Query/transform tool
• Record migration & balancing tools
 Contributing
• Post to the mailing list
http://groups.google.com/group/hivedb-dev

• Comment on our site

http://www.hivedb.org

• File a bug
http://hivedb.lighthouseapp.com

• Submit a patch / pull request

git clone git://github.com/britt/hivedb.git
 Photo Credits
• http://www.flickr.com/photos/7362313@N07/1240245941/sizes/o

• http://www.flickr.com/photos/99287245@N00/2229322675/sizes/o