Intel White Paper on Aerospike and Intel Petabyte Scale Benchmark Performance Testing in AWS Cloud
Summary
Aerospike is a high-performance distributed NoSQL platform engineered for extreme scalability at petabyte levels. This technical manual features a comprehensive benchmark study with Intel and AWS, detailing operational workloads, throughput/latencies, and significant TCO comparisons against traditional database systems. It is essential reading for IT leaders in industries like AdTech, finance, and telecommunications who require sub-millisecond latency, five-nines uptime, and massive data capacity without excessive infrastructure costs.
Page 1 Text Content
Benchmark
Running Operational Workloads with Aerospike at Petabyte Scale in the Cloud on 20 Nodes
An Aerospike white paper developed in collaboration with Intel and Amazon Web Services
August 2021
Page Summary Contents For Intel White Paper on Aerospike and Intel Petabyte Scale Benchmark Performance Testing in AWS Cloud
Page 1
Benchmark Running Operational Workloads with Aerospike at Petabyte Scale in the Cloud on 20 Nodes An Aerospike white paper developed in collaboration with Intel and Amazon Web Services August 2021
Page 2
Contents Executive overview 3 Scalability in practice 5 Petabyte benchmark 9 Workload description 9 User profile and campaign data 10 Benchmark operations and configuration overview 10 System configur...
Page 3
Executive overview Massive database scalability without massive server footprints? High availability without high maintenance? Exceptional performance without constant tuning? And total ownership cost...
Page 4
Aerospike. Finally, a leading global brokerage firm offloaded work from its mainframe DBMS and replaced its caching platform with Aerospike to improve data access speeds while more than tripling its d...
Page 5
grow from 20 to 26 nodes over a 3-year period. Compare that to Cassandra’s, which would grow from 514 to 792 nodes. Which would you rather manage? Figure 2: Server footprint for Aerospike, Cassandra o...
Page 6
Many open source and commercial solutions simply can’t scale up to petabyte-level processing for mixed workloads without critical shortcomings surfacing in one or more essential areas. For example, re...
Page 7
“Before Aerospike, we were spending more and more of our time on the care and feeding of Cassandra, and less and less time on the building of new product offerings. With Aerospike, we’ve now cleared t...
Page 8
optimized to exploit modern hardware, including multi-core processors with non-uniform memory access (NUMA), non-volatile memory extended (NVMe) Flash drives, persistent memory (PMem), network applica...
Page 9
Petabyte benchmark Confident in its ability to deliver cost-efficient scalability and high performance for operational workloads, Aerospike put itself to a test, benchmarking its server platform on a ...
Page 10
information, and non-stop global services. Perhaps that sounds familiar. Indeed, firms in many industries are facing such pressures, which is why Aerospike developed its benchmark workload to reflect ...
Page 11
Aerospike also benchmarked a 50/50 mix of read/write operations against a campaign database with 6 billion unique records totaling 1.5 TB uncompressed; with replication, the total uncompressed size wa...
Page 12
Figure 4: Overview of Aerospike’s petabyte benchmark configuration. (Note: Architecture can be readily configured to span multiple AZ’s for additional resilience.) The Aerospike server was configured ...
Page 13
User profile Read only 5,009,980 reads 100% (1 PB unique, uncompressed) Campaign 50/50 95,420 reads 100% (1.5 TB unique, uncompressed) read/write 95,420 writes 100% User profile 80/20 3,017,340 reads ...
Page 14
Cassandra solution architect to specify a configuration for Apache Cassandra 3.11.3 on AWS capable of managing the same workload for this benchmark, again assuming a 15% year-to-year growth rate. Ther...
Page 15
Infrastructure ($ USD) Fully burdened maintenance & $462,600 $574,200 $712,800 $1,749,600 $18,000 $19,800 $23,400 $61,200 support ($ USD) ($ Million USD) Table 2: Comparative TCO summary for the p...
Page 16
Figure 5: Total operational expenses by year for Cassandra and Aerospike For further details about this TCO analysis, including specifics about the Cassandra configuration, see Appendix B: Comparative...
Page 17
Signal An identity resolution firm, Signal faced latency problems with its existing Cassandra system as volumes grew. The platform’s footprint spanned more than 450 servers and further server sprawl w...
Page 18
utilization needed relative to the size of the data that we were working on. And so in order to get to the throughput that we needed, we needed to scale the number of machines to a high number of mach...
Page 19
Initially, the firm selected Aerospike to cache customer data spread across a mainframe DB2 system, a Teradata data warehouse, Microsoft SQL Server, and flat files. The company needed ultra-fast read ...
Page 20
replace its current system-of-record (managing multiple petabytes of data) and its user data profile store tracking data from its 2 billion customers. The system of record supports a multi- tenant app...
Page 21
Aerospike’s fast, predictable performance on a small server footprint makes petabyte-scale processing of real-time operational data pragmatic for a wide range of firms – not just those with extraordin...
Page 22
Appendix A: Benchmark details The petabyte-scale benchmark described earlier in this paper featured the following setup: Aerospike Enterprise Server 5.4.0.2-1 on Redhat 7: ● 20 nodes of Amazon EC2 i3e...
Page 23
● Bash scripting to encapsulate Saltstack requests Furthermore, the benchmark featured two Aerospike namespaces (databases), configured as follows: User profile namespace: ● All Flash (index and data ...
Page 24
Appendix B: Comparative TCO details To help you understand the TCO analysis summarized earlier, this appendix details the Cassandra configuration used for the comparison and the rationale for its sele...
Page 25
Object Storage size (bytes) 2457.6 2457.6 2457.6 Read peak (ktps) 5095.0 5859.3 6738.1 Writes peak (ktps) 95.4 109.7 126.2 RAM per Node (GB) 122.0 122.0 122.0 Replication size (TB) 3393.0 3901.9 4487....
Page 26
the Aerospike benchmark, but we chose to operate under the assumption that this was feasible, again to keep costs down and present an optimistic Cassandra scenario. ● Cassandra is a quorum-based syste...
Page 27
# Data Centers 1.0 1.0 1.0 Total Servers 20.0 22.0 26.0 Unique Total Data footprint (TB) 994.0 1143.1 1314.6 Data Capacity per node (TB) 60.0 60.0 60.0 Replication Factor 2.0 2.0 2.0 # of Keys (Bn) 50...
Page 28
prefer bare metal instances can customize their hardware configurations in various ways to optimize the TCO even further. Comparative cost calculations The Cassandra and Aerospike configurations prese...
Page 29
Total 3YR cost (million $USD) Table 5: Cassandra 3-year TCO breakdown Aerospike TCO - Calculation Year 1 ($USD) Year 2 ($USD) Year 3 ($USD) Infrastructure Cost Annual Server Lease Maintenance Cost IT ...
Page 30
Total 3YR cost (million $USD) Table 6: Aerospike 3-year TCO breakdown Running Operational Workloads with Aerospike at Petabyte Scale in the Cloud on 20 Nodes
Page 31
About Aerospike The Aerospike Real-time Data Platform enables organizations to act instantly across billions of transactions while reducing server footprint by up to 80 percent. The Aerospike multi-cl...
Manual Details
| Brand | Intel |
|---|---|
| Pages | 31 |
| File Size | 1010.80 KB |
| Published | June 03, 2026 |
65 views
Enter the captcha to get the download link:
Frequently Asked Questions
What is the maximum performance achievable at a petabyte scale?
The platform delivered 4 - 5 million transactions per second (TPS) for read-only and mixed workloads.
How can Aerospike improve operational costs compared to alternatives like Cassandra?
For a 3-year period, the TCO estimated for an Aerospike cluster was significantly lower ($4.17 million) than that of an Apache Cassandra-based architecture ($15.29 million).
What level of data uptime and consistency does Aerospike provide?
It maintains five-nines uptime with globally distributed, strongly consistent data.
Does the platform reduce server footprint when handling massive data volumes?
Yes, it reduces the server footprint by up to 80 percent.