Dmitriy Setrakyan, founder and Chief Product Officer at GridGain, delivered this talk during the April 11 Bay Area In-Memory Computing Meetup. Abstract: The 10x growth of transaction volumes, 50x growth in data volumes -- along with the drive for real-time visibility and responsiveness over the last decade -- have pushed traditional technologies including databases beyond their limits. Your choices are either buy expensive hardware to accelerate the wrong architecture, or do what other companies have started to do and invest in technologies being used for modern hybrid transactional/analytical processing (HTAP). This presentation covered: * The requirements for real-time, high volume HTAP * Architectural best practices, including how in-memory computing fits in and has eliminated tradeoffs between consistency, speed and scale * A detailed comparison of Apache Ignite and GridGain® for HTAP

1. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Comparing Apache Ignite and Cassandra
For HTAP Applications
Dmitriy Setrakyan
Apache Ignite PMC Chair
GridGain Product Management

2. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Agenda
• Ignite vs. Cassandra
– Denormalized Architecture or Collocated Processing?
– Row-level Isolation or Distributed Transactions?
– Benchmarks
• Q & A

3. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Memory-centric distributed
database, caching, and processing platform
Distributed database built for
performance, scale, and availability

4. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Ignite vs. Cassandra:
Denormalized Architecture or Collocated Processing?

5. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Cassandra Path: Denormalized Architecture
• Denormalization Strategy
– Increasing Performance in Favor of Data Redundancy
• Pros
– Performance
• Cons
– Query-Driven Architecture
– Easy to Start and Hard to Evolve

6. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Let’s Take an Example

7. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Start With a Question!
Q1: What are the car models produced by a vendor within a
particular time frame (newest first)?

8. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Run Queries
Q1: What are the car models produced by a vendor within a
particular time frame (newest first)?
Extra Queries Supported Out of the Box

9. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Somewhere in Production…
Q2: What is the number of cars of a specific model produced
by a vendor?

10. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Let’s reuse existing table!
Q2: What is the number of cars of a specific model produced
by a vendor?

11. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Let’s reuse existing table!
Q2: What is the number of cars of a specific model produced
by a vendor?
InvalidRequest: code=2200 [Invalid query]
message="PRIMARY KEY column
"car_model" cannot be restricted (preceding
column "production_year" is not restricted)"

12. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Solution: Create New Table
Q1:
Q2:

13. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Ignite Path: Affinity Collocation
• Data-to-Data Collocation
– Countries and Cities, Vendors and Cars
– Efficient SQL JOINS
• Compute-to-Data Collocation
– Collocate compute and data
– Reduced Network Traffic -> Better Performance

14. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Collocated Data Distribution

15. © 2018 GridGain Systems, Inc. GridGain Company Confidential
1. Initial Query
2. Query execution over local data
3. Reduce multiple results in one
1
2
23
Collocated JOINs

16. © 2018 GridGain Systems, Inc. GridGain Company Confidential
DURABLE MEMORY
DURABLE MEMORY
Ignite Cluster
F2, C2, M2
F = F1 + F2
C = C1 + C2
Collocated Computation
Biological Evolution
Simulation
Chromosome and Genes Cluster
M = M1 + M2
F1, C1, M1
F = Fitness Calculation
C = Crossover
M = Mutation
Machine Learning: Genetic Algorithms

17. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Ignite vs. Cassandra:
Row-level Isolation or Distributed Transactions?

18. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Cassandra Path: Lightweight Transactions
• Cassandra is Eventually Consistent DB
– With Tunable Consistency
• Lightweight Transactions
– Aka. Compare and Set Transactions
– Row-Level Isolation
• Applicability
– Atomic and linearizable updates of a record
– Prevent duplications on inserts

19. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Ignite Path: ACID Transactions
• Distributed ACID Transactions
– Pessimistic/Optimistic
• 2 Phase Commit
– From RAM to disk
• Deadlock-free Transactions
• Applicability
– No limitations. Classic Transactions.

20. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Ignite vs. Cassandra:
YCSB Benchmarks

21. © 2018 GridGain Systems, Inc. GridGain Company Confidential
YCSB: Average Load

22. © 2018 GridGain Systems, Inc. GridGain Company Confidential
YCSB: Ramping Up Load

23. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Caching Cassandra With Ignite
• No rip-and-replace
– Keep Cassandra
• Automatic Read/Write-Through
– Key-Value Only
• Distributed SQL
– Over Ignite Data
• ACID Transactions
– Ignite layer

24. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Summary: Ignite or Cassandra?
• Simplified Architecture
– Denormalization vs Affinity Collocation
• Data Consistency and Transactions
– Ignite: ACID Transactions
– Cassandra: Row-level Isolation
• In-Memory Store and Performance
– Ignite: read intensive and mixed workloads
– Cassandra: write intensive workloads (big load)

25. © 2018 GridGain Systems, Inc. GridGain Company Confidential
Thank you for joining us. Follow the conversation.
https://ignite.apache.org
Any Questions?
#apacheignite
#gridgain
#dmagda
https://www.gridgain.com