Many IT managers evaluating flash for data center workloads are comparing between all-flash, all-hard drive (HDD) or hybrid (flash and HDD) array options. But flash is not a “one size fits all” solution, particularly not in the case of data center workloads where I/O characteristics and endurance requirements differ between various applications.
Enterprise-grade flash solutions, such as SAS, SATA, PCIe and Flash on DIMM solutions in SanDisk®’s portfolio are architected to support the mission-critical nature of enterprise applications with extreme performance, superior reliability, optimal endurance and full data protection. Obviously, these features come at an added cost.
In the case of hyperscale data centers, deploying petabytes upon petabytes of data, flash has been enabling scalability at lower costs by delivering higher performance at just a fraction of the hardware, electricity and cooling costs when compared to HDDs. What few people know is that hyperscale data centers have been taking advantage of Client SSDs in their environment to achieve even better cost efficiencies for flash at scale.
Client vs. Enterprise SSDs
What differentiates client, Ultra-Value SSDs from enterprise SSDs? The key differences between these two categories of SSDs lie in their essential features – the maximum read/write performance (particularly the number of writes per second), data access latency, maximum endurance, level of reliability and data protection mechanism (for example, client SSDs do not have a capacitor for power fail protection as enterprise-grade SSDs do).
The other key differentiator is cost. Lacking these key features, client SSDs come at a lower price point. For hyperscale data centers, deploying thousands, or even hundreds of thousands of SSDs, the difference between deploying client SSDs vs. Enterprise-grade SSDs can add up to millions of dollars.
Hyperscale Environments Are Different
Hyperscale data center environments are different than enterprise data center ones. It’s not only sheer scale that makes these data centers different but the fact that these computing environments run particular workloads, unlike enterprises that host an array of different applications for various departments and users. In addition, they often deploy custom designed hardware and invest in in-house engineering knowledge in order to maximize infrastructure efficiencies at massive scale.
Ultra-Value SSDs require good management capabilities. Hyperscale customers have the software and hardware knowhow to architect applications in a way that can overcome the vulnerabilities of client and Ultra-Value SSDs. This is not the same in the case of enterprise data centers, where most organizations do not have the expertise, personnel, or time to do this type of management.
For this reason, in most enterprise environments, Ultra-Value SSDs are an unsuitable solution. Furthermore, without proper management, using low-endurance SSDs can end up being more costly due to frequent replacement. (I recommend reading Hemant Gaidhani’s blog post on endurance and Total Cost of Ownership to better understand the impact of endurance on achieving the best return on investment.)
Client SSDs: Hyperscale Workloads
Client SSDs are most beneficial for workloads that are predominantly read intensive and don’t require high write I/O or endurance. Here are some examples of how our customers deploy Ultra-Value SSDs:
Video on demand
A very popular service that most of us take advantage of is video on demand. Video is a heavy application that requires massive infrastructure to deliver a seamless, high-quality and stutter-free user experience (I think we can agree that there is nothing more frustrating than having a video load slower than it streams.) Hard disk storage is a serious bottleneck for video workloads due to limited scalability and throughput degradation when a large number of simultaneous users and streams are required.
Typically, video on demand services place data (video) on storage once a day, or several times a week. Users then read the same data over and over, streaming from the same device. The read/write ratio is usually around 99% read and 1% writes, with a maximum of 95% read / 5% write ratio for these scenarios, making it a great fit for Ultra-Value SSDs.
The way in which the data is written to storage is also decisive in whether your workloads are suitable for Ultra-Value SSDs. If your data generally requires small, random writes, enterprise SSDs are a far more fitting solution. But most hyperscale customers are managing the data write process. Video in particular is a great example of a large data transfers, as well as sequential transfers, that can help to overcome endurance vulnerabilities. Furthermore, the ability to manage when data is uploaded and written to the storage (e.g. avoiding ‘rush hour’ when majority of users are online) is key for a successful storage strategy.
Search Engines
Another common use for Ultra-Value SSDs is caching and indexing of search engines to accelerate the retrieval of information and achieve better reliability. Retrieving random data is the Achilles Heel of spinning media, as HDDs need to physically seek data on disk. As the physical location of data is irrelevant with flash, it can deliver consistent read performance with no delays and far lower latency.
In this particular workload, keywords are written into tables that are indexed and cached on SSDs. Those keywords are used as pointers to data that is typically located in another layer of storage. Index updates don’t happen every second, not even every minute, so the write cycles are kept relatively low (generally a 90% read / 10% write characteristic) and can be done in a sequential operation. This makes this workload a good fit for Client SSDs.
For data protection and uninterrupted service, one can backup all the data by putting the drives into a RAID configuration (usually a software RAID). Should one drive fail, data backed up in another system duplicates into the same layer. In this manner, the lower endurance of Ultra-Value SSDs can be counterbalanced.
Are Ultra-Value SSDs Right for You?
Ultra-Value SSDs are making a big impact in hyperscale workloads, delivering flash at scale at lower costs. The use case is a unique one that requires good software management and in-depth knowledge of the storage system and data path, to overcome vulnerabilities.
SanDisk X300DC is a high-performance SATA drive designed to deliver consistent read performance with low active power and high quality of service (QoS). It is intended for customers with data centers who are already using client drives, require entry-level storage solutions, or seek a cost effective solution to replace their hard disk drives and do not require enterprise-level endurance, reliability, or power failure protection (PFP).
If you’re wondering which flash solution best fits your workloads, talk with our experts; we’ll help you define the right strategy for your data center.
