By Steve Sarsfield
It’s common to see technologies running faithfully in companies for years, even decades. Like a history lesson, you might see mainframes capturing data, phone systems that haven’t been touched in years, and even analytical databases that have been running since the days of dial-up modems and flip phones.
However, innovation is happening today at a breakneck speed in both the hardware being shipped and the efficient software being written to optimize every business function. Companies across the most data-intensive use cases, such as telco, ad tech, and government, should consider how to make their current legacy data analytics solutions, specifically on-premises infrastructures, more modern and more efficient.
For example, in 2015, large enterprises commonly operated data warehouses in the 1–10 TB range (though very large organizations sometimes managed 10–100 TB). Today, it is more and more common to see petabyte and multiple petabyte data stores. Solutions like Netezza, Exadata, or Vertica that once dazzled predecessors as specialized analytical powerhouses, today need to be re-evaluated on how they perform, manage costs, and achieve optimal efficiency.
This blog post looks to examine the top nine biggest breakthroughs driving innovation in on-prem analytics. In particular, I’ll take a closer look at 4 breakthroughs with on-prem solutions that are being used to analyze the most complex, compute intensive workloads.
Ten or fifteen years ago, software performance improved largely through higher CPU clock speeds, which rose from sub-gigahertz to over three gigahertz. However, power and thermal limits eventually slowed this trend. Manufacturers shifted to multi-core designs instead. Today, they can squeeze way more cores into your processors than ten years ago. This is an ideal situation for data analytics platforms that rely heavily on scanning, filtering, aggregating, and joining large datasets.
From the software perspective, modern engines have the advantage of being designed to partition data and tasks from the outset to be processed concurrently. Legacy on-prem databases are stuck. Systems built with single-thread assumptions can’t automatically distribute these operations across multiple cores. Retrofitting legacy code for parallelism is non-trivial because data structures, memory management, and scheduling often need fundamental changes. Legacy analytical software written for fewer cores can’t just “drop in” parallelism. A major refactoring is necessary to exploit modern CPUs like AMD® EPYC fully.
As to value, using all your cores means that you can handle more data, use fewer servers, lower electrical costs, and provide a greener way to process data.
We’re also discussing cores and their impact on carbon emissions and power use. Modern processors deliver higher performance per watt. Semiconductor manufacturing processes (e.g., 14nm → 7nm → 5nm and beyond) have steadily lowered the wattage needed to run the core. Modern many-core CPUs benefit from these process advancements, squeezing more logic (cores) onto a die while keeping power within practical limits.
In highly parallel workloads (analytics, databases, AI/ML), many moderate-speed cores can deliver substantially higher throughput at a lower total wattage than one or two fast cores. If the analytics platform can fully use the cores, the data center can now handle larger analytics workloads without a linear increase in power consumption. The data center is more cost-effective. Modern processors handle more work per server, allowing organizations to reduce the footprint of on-prem infrastructures.
Input/Output Operations Per Second, or IOPS, is king for databases. Databases need to handle data to be stored and accessed without delay, and IOPS measures how fast they can do so. SSDs offer dramatically lower latency and higher throughput compared to mechanical drives.
Workloads that involve huge read/write operations, like real-time analytics, benefit tremendously from SSD-based systems. Imagine having a streaming data workload requiring 10,000 IOPS to maintain high performance. If you’re looking at handling machine logs or leveraging IoT data, it’s common. To achieve the IOPS, you need a minimum of 40 hard drives in tandem (40 X 250 IOPS) to store that data away, or you can use one SSD and still have room for expansion. It’s a huge difference in the amount of hardware you need.
Like the other innovations in this list, it’s not just a matter of switching to new hardware.
While switching to SSDs might have some performance benefits in legacy databases, the breakthrough comes when the analytical platform can hotwire the OS and use new pathways specified by NVMe standards that bypass the operating system.
The jump from rotating disk to SSD greatly boosts I/O performance, shrinking data-access bottlenecks and enabling near real-time analytics for large datasets, even on prem.
Complexity used to be the challenge with on-prem systems. Ten years ago, it was common for companies to use multiple solutions to handle their data pipelines. This meant gaining expertise, licensing, and maintaining various apps. An IT team might string together Redis, Apache Cassandra, or Apache SPARK for streaming data and combine it with Postgres or Teradata for OLAP workloads.
New on-prem solutions offer a complete solution that combines data ingestion and integration, orchestration, transformation, and enrichment with the ability to analyze real-time data. These data pipeline systems are often part of a managed service that provides some of the simplicity of the public cloud but in an on-prem deployment.
Combining real-time analytics (RTA) and OLAP in a single platform cuts operational overhead by eliminating multiple, disjointed systems. You reduce licenses, vendor relationships, and maintenance, and your data teams benefit from a unified skill set. Centralized governance and security further mitigate duplication and inconsistencies. The result is an efficient environment that delivers real-time and historical insights while lowering costs and complexity.
Innovation Rapid-fire
Let’s quickly review the rest of the innovations that are boosting on-prem analytics.
How Ocient Modernizes On-Prem Analytics
Ocient was built to use every ounce of CPU parallelism possible. Modern HPC-inspired architectures say, “Give me all your cores, and I’ll put them to work.” When your platform is deeply threaded and built from the ground up for parallel processing, you see impressive query performance that scales linearly (or near-linearly) with every additional core.
When storing data, Ocient leverages Compute Adjacent Storage Architecture® (CASA), which ties together NVMe drive storage with the system compute resources to optimize performance. This design keeps records near and accessible for computation and avoids many common bottlenecks for database engines. The design achieves superior query performance when it operates on trillions of rows of data.
Ocient leverages the advanced core density and power efficiencies of 4th Gen AMD EPYC processors, enabling organizations deploying Ocient to achieve 3X better performance for compute-intensive workloads. With up to 192 cores, the new processors reduce operational costs through improved power and energy efficiency. They deliver about a 3X reduction in power consumption per core.
An example of a company at the forefront of leveraging these new hardware innovations is Basis, a marketing and advertising technology provider. Basis has replaced multiple legacy data pipelines and consolidated data operations with new multicore-capable software while storing data on SSDs. Basis can ingest and query billions of rows of information at faster speeds. This shift to a modern, cloud-optimized analytics solution reduces hardware, maintenance, and resource requirements, enabling Basis to decrease overall data management expenses. The use case discusses reduced infrastructure costs, decreased operational overhead and faster time to insights.