Avoid Tricks and Treat Yourself to Energy-efficient Storage

In the world of enterprise storage, sustainability has become a critical consideration, but comparing energy efficiency across storage products is far more nuanced than a simple benchmark or a certification test score. While certain benchmarks might claim to identify the most energy-efficient system, the reality is that such tests often fail to capture the full range of factors that impact real-world energy consumption.

One Size Does Not Fit All

Storage systems offer multiple configuration options-drive sizes, cache, and data protection schemes-which directly impact power and space consumption. A single synthetic test result does not fully capture these variations. For example, a storage array configured with a large cache size can skew results in favor of systems that prioritize performance from cache, a tactic that doesn't necessarily translate to practical energy efficiency in the real world.

Many of these configurations are customized based on specific customer requirements. A storage system with larger or more efficient drives, more advanced data protection mechanisms, or more substantial caching capabilities will consume energy differently than one with smaller drives and minimal data protection. Therefore, a meaningful comparison of energy efficiency requires normalization around the specific configurations that would meet a customer's actual needs.

Caution Ahead: Performance Benchmarks Disguised as Energy Efficiency Metrics

Performance benchmark tests like the Trans Optimal Point Hot Band Workload Test (TOPHBWT) measure IOPS per watt based on a specific, limited-duration synthetic workload. While such a metric can highlight peak efficiency in a particular scenario, it doesn't account for the various workload configurations and operational demands that different customers may face. Therefore, customers should be cautious when interpreting such results, as they may not reflect the actual energy efficiency of a storage system when configured to meet their specific needs.

It's also significant to note that performance benchmarks can be heavily influenced depending on how much test data they permit to be cached, a testing artifact from the hard disk drive (HDD)-based era. When data is retrieved from cache it speeds up I/O and boosts the overall benchmark score. The reality is application data is not always available in cache and must be retrieved from where it resides over longer periods of time-the storage media. This means the true energy consumption and performance impact of retrieving data is not completely captured by benchmarks like TOPHBWT. As a result, systems that can scale to very high "hero" IOPS numbers regardless of cost, configuration, and overall energy consumption will appear to be more energy efficient than those that cannot. But before accepting the "Highest IOPS = Highest Energy Efficiency" approach, let's delve into it a bit more.

Here's a simplified example showing a data center storage solution that can scale the number of controllers and/or the number of expansion shelves to scale performance and capacity:

Table 1: Annual energy consumption as array controllers or expansion shelves are added to meet capacity requirements.

Table 1 illustrates how IOPS per watt remains constant across array configurations with 2 to 10 controllers, assuming no expansion shelves. However, annual energy consumption differs significantly. Comparing the 2 controller with 8 expansion shelves solution to the 10 controller no expansion shelf solution using the corresponding IOPS/Watt values suggests the 10 controller configuration is the most energy efficient even though it uses nearly double the energy! Quite an interesting trick if you ask us.

In this example, the required storage capacity, 4.8PB, exceeds what all but a 10-controller system can contain in just its controller enclosures. (At least some expansion shelves are required in the other examples in Table 1.) The second line (in Table 1) for each controller quantity illustrates the outcome when just expansion shelves are used to meet the capacity requirement. For storage systems that scale the number of controllers along with capacity, IOPS/Watt is greatest for configurations with fewer or no expansion shelves. Despite the fact that the expansion shelves themselves consume less than 25% of the power of two additional controllers, they negatively impact the value of the IOPS/Watt metric. This dynamic incentivizes vendors to scale capacity by adding controllers rather than expansion shelves to achieve the highest IOPS/Watt, even when additional IOPS are not required. That unnecessarily increases both cost and energy consumption, effectively assuming that the total energy consumption of the solution does not matter. In fact, it does matter and the requirement to add controllers as capacity is expanded can lead customers to purchase wasteful, energy-inefficient solutions with far higher energy consumption than would otherwise be required.

This is why Pure Storage believes in a more comprehensive and transparent approach to storage solution evaluation with our customers-one that establishes workload characteristics and maximum performance requirements first and then examines energy and storage efficiency across various solution options.

For most enterprise storage purchases, customers care about the trade-offs between a number of different measures: performance, capacity, energy consumption, cost, ease of use, and likely a few others. At Pure Storage, we believe enterprises should use simple and transparent metrics to evaluate storage solutions. Using metrics such as storage capacity per watt (TB/watt), storage density per rack unit (TB/RU), watts per maximum throughput bandwidth (watts/GB throughput), and $/effective capacity ensures that each configuration meets performance, scalability, and total cost of ownership objectives. An example of how much clearer efficiency comparisons become can be seen in the TB/Watt column in Table 1. The configurations that use the least amount of energy annually and meet the performance and capacity requirements have the highest value. We are convinced that systems designed to use flash like flash instead of like disk will always be more efficient in terms of energy consumption and rack space utilization.

The energy and space efficiency advantages of Pure Storage center around the fact that we manage flash as flash and can deploy storage devices that will soon be up to five times larger in volume than the largest commodity SSDs that are shipping today but consume roughly the same amount of power (on a per device basis). We've been shipping 75TB DirectFlash® Modules (DFMs), which are flash storage devices of our own design, for the last two years and will be shipping 150TB DFMs starting later this year. While some storage drive vendors have announced 60TB SSDs, most storage system vendors are only shipping 15TB and 30TB SSDs in volume. This density advantage enables us to build very efficient multi-PB storage infrastructures at $/GB costs that rival comparably sized HDD-based systems. We have smaller DFM sizes as well, including 2.2TB, 4.5TB, 9.1TB, 18TB, 36TB, and 48TB, that bring many of the same benefits (in terms of how we manage the flash media for efficiency and the reliability of the devices) to smaller systems.

Evaluating Energy Efficiency

To meaningfully compare the energy efficiency of storage systems, it's essential to evaluate configurations that align with realistic applications, workloads, and use cases. What matters most is how a system performs in the specific context in which it will be deployed. For example, while a benchmark may highlight the peak efficiency of a system under a high-IOPS workload with minimal data protection, most enterprise applications require a more balanced approach-factoring in storage capacity, fault tolerance, long-term operational costs, data services, and cyber resiliency in the event of data corruption (or due to cyberattacks).

The trade-offs between performance, energy consumption, and space efficiency are vital considerations for enterprises. What may appear as an efficient solution in a benchmark may not perform as well when scaled to meet the storage services, application performance, data protection, and capacity needs of a business. Focusing on just one efficiency metric can lead to buying too much or too little to satisfy all customer needs. By ensuring that energy efficiency metrics are tied to configurations that reflect real-world deployments, customers can make more informed decisions about the best storage systems for their needs.

While IOPS/Watt benchmarking with tests like TOPHBWT can provide a glimpse into the energy efficiency of storage systems under specific conditions, they don't provide a full picture. The configurability of modern storage solutions, along with the varying needs of customers, means that energy efficiency is more complex than a single number. Normalizing performance and storage efficiency metrics to actual customer configurations offers a more reliable way to compare systems.

Enterprises must also be cautious of vendor claims based solely on "spec-sheet" comparisons or theoretical configurations that do not accurately reflect real-world deployments. Spec-sheet numbers can be carefully selected to highlight peak performance or efficiency under ideal, often impractical conditions. These comparisons frequently fail to account for how storage systems will actually be used in production environments, where total cost of ownership, performance, data protection, and scalability requirements vary dramatically. Customers should always ask whether the configuration used in a benchmark is something they would realistically deploy and how it would meet their business and application needs over time.

The Pure Storage Advantage

At Pure Storage, our approach is fundamentally different. We begin by collaborating with customers to understand their specific business and application performance requirements. Rather than pushing a one-size-fits-all solution, we focus on right-sizing the system to meet those needs, delivering not just performance but also superior energy and space efficiency. Our storage solutions are designed to use flash like flash, leveraging larger and more efficient flash modules that enable greater density and lower power consumption compared to solutions based on commodity solid-state disk and mechanical hard disk configurations. This allows our systems to consume far less energy and space than competing solutions, making them more environmentally friendly and cost-effective in the long run.

In addition to offering immediate energy savings, our systems are built to last and can be non-disruptively upgraded an unlimited number of times. They're more reliable, reducing the likelihood of failures and the associated costs of downtime or data loss. Data collected from our installed base over the last 13 years indicates that our customers are routinely achieving 10-year product life cycles with our storage systems, while our competitors routinely encourage their customers to upgrade to new products every four to five years. Furthermore, the longevity of our systems means they generate significantly less e-waste, helping to minimize the environmental impact of outdated technology being sent to landfills. By focusing on durability and sustainability, Pure Storage ensures that customers benefit not only from cutting-edge performance but also from longer product life cycles, contributing to a more sustainable and responsible IT environment.

It's crucial for enterprises to move beyond surface-level comparisons and look deeper into how storage systems align with their operational needs. Our comprehensive approach-centered around right-sizing the solution for performance, energy efficiency, space optimization, and long-term reliability-ensures that customers not only meet their business objectives but also contribute to a more sustainable future by reducing energy consumption, space usage, and e-waste.

Public link

Please use the above public link if you want to share this noodl on another website.