FAQs

Yes, install the same as any other NVMe SSD. There are no special drivers or software configuration required. It works with the system using standard NVMe commands. So, migration from old devices onto modern SSDs would follow a workflow like any other.

The drives do not require any special software or driver installation.

Encryption for data-at-rest is table-steaks for enterprise & data center drives. Offloading compression to the drives is more efficient and scalable than CPU-based compression. To enable additional compute functions to be done in or near the drives, the drives need to be able to manage the compression and the encryption.

The drives can increase capacity through NVMe commands in the user’s environment. You don’t need to pay us an additional fee to expand the capacity of the drives you bought… they’re your drives, use them however you need to!

We are happy to consult with your team to discuss the storage capacity as well as the total workload needed to help you determine the right number of drives and total capacity of storage.

We plan for quarterly updates which include release notes. Of course, if a security issue arises, we will issue an off-cycle release as soon as the fix is implemented.

Nope, you can replace a ScaleFlux drive with any other vendor’s drive… you’ll just need to buy one with 2x the capacity to store the same amount of data and you may need to buy 2 of their drives to reach the same system performance.

ScaleFlux drives work anywhere and NVMe SSD can be used and there is no minimum requirement for CPU / DRAM. However, these are enterprise drives with requirements for power, airflow, and ambient temps to avoid thermal throttling. No integration into the applications is required. To take full advantage of the extended capacity, users do need to monitor the NUSE parameter via the NVMe protocol commands.

Absolutely correct. The compression ratio is dependent on the type of data (anyone who tells you differently is misleading you!). We have a compression estimator tool to let you test your data (with or without having deployed our drives, the estimator mimics the compression algorithm in the drives). That said, we have examples of the CR customers have realized across a variety of data types.

Yes, add individual drives as you need them.

The only real infrastructure change is in the capacity monitoring. When using the compression for extended capacity, you need to monitor the physical capacity gas gauge as well. This is handled through standard NVMe commands for “thin-provisioned namespaces.” We can provide a guide to those commands upon request.

ScaleFlux SSDs are installed and present the same as any other NVMe SSD and as a result have the same requirements. They can cohabitate peacefully in any system that supports NVMe devices.

No, those are a bit beyond the scope of the SSD. We strongly recommend you continue to manage your Backup, DR, and Ransomware recovery with companies specialized in those areas.

To clarify, we are not deduplicating. We are using hardware compression engines to perform lossless, block-level compression at full NVMe data rates. Dedupe is best done on very large datasets and at the file or object level. The ScaleFlux CSDs are block storage devices that do compression on small blocks of data (4KB blocks). This granularity of compression allows for line-rate compression and decompression to meet the needs of latency-sensitive applications.

Nope! Running the ScaleFlux drives in production does not require you to use them in your Disaster Recovery (DR) or in other parts of your local infrastructure for that matter. Your DR simply needs to have enough capacity to handle your data set.

That’s a big question! We’re just at the early stage of deploying CS. As an industry, we are still working through all the standards to enable computational storage to reach its full potential. Over time, CSDs can contribute to the efficiency of connecting data to compute and memory for a more efficient and higher performing data center.

I have seen instances in which users try to use client-grade drives for server applications, setting themselves up for disappointing failure rates. Using server management tools and working with system integrators / VARs to properly set up their systems to meet their needs is a good starting point.

ScaleFlux has been the first to bring drive-based compression to market in a NVMe SSD. Enabling the full benefit of compression to expose extra capacity in the drive required a significant enhancement to the controller design and to the firmware architecture. Yes, a couple of other vendors have brought compression into the SSDs recently. Storage array vendors have deployed host-based compression (taxing the CPUs, low throughput, not highly scalable) for many years.

We are seeing many companies repatriate workloads from the cloud to on-prem! Not every workload is well suited to cloud – or is economical to run on the cloud.

Workstations are possible. Computational storage drives are industry standard form factors (U.2 today, others coming). If the workstation can provide ample airflow for an enterprise NVMe SSD, it’s an option. We do have some customers looking into using the drives to improve performance and local capacity for workstations.

Cloud storage is an architectural choice typically requiring you to migrate your data to a 3rd party’s infrastructure. Computational Storage Drives can be used by cloud service providers to reduce their costs of delivering their service (which may or may not get passed on to their customers) or by companies with their own on-prem or co-lo infrastructure.

Computational storage refers to offloading storage-centric computing tasks from the CPU to a device that’s closer to the storage media. In the case of ScaleFlux, we are talking about Computational Storage Drives – drives with built-in compute or data-processing functions.

SSDs from ScaleFlux are natively NVMe, using the standard NVMe drivers available with any OS. There is no host-awareness requirement. All the benefits are transparent, so the host is not even aware. This is the same for applications. You get the benefits without any software changes or reconfiguration.

I have seen instances in which users try to use client-grade drives for server applications, setting themselves up for disappointing failure rates. Using server management tools and working with system integrators / VARs to properly set up their systems to meet their needs is a good starting point.

Monitor the computational storage standards at www.snia.org

Typically, the SAN array vendor will select any replacement drives.

ScaleFlux SSDs can plug into any U.2 or U.3 drive bay. Other form factors are in development.

There are no OS limitations as we use the standard NVMe drivers.

That’s pretty much the flow! The drive receives a write IO, performs compression and collation to create page-aligned blocks of data, encrypts the data and then writes it. The process is reversed on reads (decrypt, decompress, send to host).

Yes, ScaleFlux has a utility you can download to estimate your data’s compressibility. https://github.com/kpmckay/compression-estimator

No, we use industry standard compression algorithms.

The drives are block devices. A large file sent to the drive gets divided up into 4KB blocks for compression (invisible to the user) and written to the NAND. RAW formant video / images are compressible. Drone imaging that has already been compressed with a video or image compression codec will not see much residual compression.

The CSD 3000 uses hardware engines to perform the compression/decompression function.

When compression is done in hardware on the drive, compression becomes a performance accelerator as it minimizes SSD overhead activities such as garbage collection. The result is that write amplification is reduced and latency is improved especially in mixed read/write scenarios as the amount of traffic into and out of the flash memory chips is greatly reduced. Also, all of this is offloaded from the CPU so that it can spend more cycles doing higher value works.

While both ScaleFlux and Pure provide Flash-based solutions, the two companies are addressing different parts of the storage market. Pure sells All Flash Arrays (AFA) – complete systems with their own proprietary software and management tools. ScaleFlux sells drives for use in any server OEM’s systems. ScaleFlux drives do not require any proprietary drivers or new software to install. The ScaleFlux CSD (computational storage drive) plugs into the same slots that other vendors NVMe SSDs plug into. Both companies offer data reduction in their solutions. Pure manages compression and deduplication at the system level, much like other storage array vendors. At the array level, the focus is first & foremost on maximizing the data reduction, which typically aligns with compressing large chunks of data. The coarse-grain compression is not appropriate for high-performance applications as it introduces significant read amplification and latency penalties. ScaleFlux brings compression down into the individual drives and compresses small blocks of data. The fine-grain compression enables high-performance applications with mixed read-write workloads to utilize compression as an accelerator as well as a storage cost savings tool.

For server OEMs, we have tested systems from several vendors and work with system integrators and value-added resellers (VARs) to deliver systems with our drives to customers.

Yes, since each drive includes compression engines specifically designed to handle the per-drive data rates, you can scale your throughput linearly with each incremental drive you add. You’ll be distributing and parallelizing that task across the drives.

ScaleFlux frees up CPU cycles for your applications by taking on the burden of compression & decompression. With consistently lower response times from the drives, users can realize an increase in CPU efficiency (less time lost to “wait” cycles)

The ScaleFlux CSDs enable users to cache larger volumes of data in the compute nodes to reduce the need to thrash the local cache (which in turn creates more network traffic)

Applications and servers see performance gains with ScaleFlux drives vs. other enterprise NVMe SSDs. By offloading the CPU from storage processing and optimizing flash memory to reduce write amplification, latency can be greatly improved, and bottlenecks alleviated by keeping traffic off the bus. By using dedicated hardware on the drive compression can be processed nearly 100x faster than in the CPU. The latency and performance improvements are direct gains from doing hardware-based compression in the SSD controller ASIC. The benefits for network traffic are more second-order improvements – by increasing the effective storage space in the server, the server can reduce the number of times it needs to fetch data from across the network, alleviating network traffic.

In mixed read/write workloads such as OLTP, customers report 2-4x increase in transactions per second. Since the CSD’s do not use CPU resources or host DRAM to manage compression, performance scales with each drive you add… until you max out how many database transactions the CPU can handle of course!

In any workload involving a mix of read and write traffic, you can see 2x or more the performance of ordinary NVMe SSDs.

In the 4th generation of drives with our PCIe 5 ASIC, expect performance to be approximately 2x on all of the performance metrics. The current gen ship with up to 16TB of physical capacity and supports up to 24TB of data storage. Next gen includes plans for up to 32TB physical and 64TB of data storage, though higher capacities are possible as NAND densities increase.

Potentially in the future. The current drives are intended for servers.

No, though when data is incompressible (e.g., pre-encrypted), we won’t be able to leverage compression to improve performance and QoS. If you suddenly switch from sending compressible data to sending incompressible data, the write performance will eventually decline as the NAND on the drive fills. However, it’s the average compressibility of the data that will influence performance more than short-term shifts.

ScaleFlux carefully designed its controller ASIC and the drives to comply with the power constraints of the enterprise SSD form factors. By integrating the hardware compute engines directly into our ASIC, we avoid wasting board space or drive power moving data between multiple components. ScaleFlux drives typically operate within ~1W of other high-performance NVMe SSDs… but we get 2-4x the performance out of that extra 1W, making us vastly more efficient in terms of performance per watt.

Yes! The CSDs must meet the industry specs for form factors & power constraints.

ScaleFlux drives include power loss data protection circuitry and in-drive capacitance to transfer volatile caches to non-volatile storage in the event of surprise power loss.

Thanks! That’s a big part of the value – reducing the number of systems & components to achieve a given workload.

We do not have a separate fee for ongoing maintenance.

The drives carry a 5 year / Total-bytes Written warranty which is the common practice for enterprise SSDs. With higher compression ratios (CR), the effective endurance (TBW) of the drives increases dramatically. With CR as low as 1.2:1, users see 2x the TBW of other vendors drives. With 4:1 CR, that can reach 9x the endurance!

Yes, we have a couple of TCO calculators. Please contact us to discuss your situation.

We do not charge licensing fees. We sell you the drives. If you choose to utilize the capacity multiplier feature to store more data, you don’t have to pay us anything additional.

ScaleFlux drives are competitive with high-performance TLC drives of similar physical capacities.

ScaleFlux drives reduce the cost per GB of data stored. Pricing is competitive with drives of similar physical capacity.

They are designed for servers with controlled environments and significant airflows. Some workstations can be viable as well. These are not suited for commodity PCs or laptops.

Yes, we do support on-site testing of the products for qualified customers. Contact us to arrange a consultation. https://scaleflux.com/contact-us/

ScaleFlux works with partners to deliver its drives in partners’ storage appliances and arrays.

Yes, we can provide on-site support.

Yes, we are based in Milpitas, CA.

We can replace drives within 48hrs for RMAs.

At this time, we’ve kept it simple for users and pre-set the computational storage functions. We are engaging with partners in the enablement of customized functions. If you are looking to develop custom functions, please contact us. https://scaleflux.com/contact-us/

The ScaleFlux CSD 3000 is a third-generation computational storage SSD built with a custom ASIC around flash controller and an eight-core Arm processor. We built these NVMe SSDs for enterprise servers and the demanding workloads & reliability specs of enterprise customers. We don’t have a retail or consumer version.

In the current generation of CSD (computational storage drive), we have implemented compression in hardware (vastly more power & performance efficient than using CPUs). To get all the benefits of compression in the SSD, we have to develop a novel flash management process to manage the variable lengths of data that result from compression (vs the fixed length blocks that are typically managed by SSD controllers).

The drives are warranted for 5yrs, or lifetime petabytes written (PBW) (which scales with capacity). You may also see Drive Writes Per Day (DWPD) as the indicator of how many times you can fully overwrite the entire capacity of the drives each day. Given even a very modest amount of compression (1.2:1 compression ratio), the drives achieve 2x the endurance of other NVMe SSDs.

Our current generation of products have a custom SoC ASIC designed by us, we do not use FPGA. This is to reduce costs, improve performance, and to enable NVMe compatibility.

We have manufacturing sites both in North America and in Asia.

Our drives are 100% NVMe compatible. We are the only computational storage product that requires no special software or drivers.

No, we use the same SKU for both. In your latency-sensitive application, you may choose to maximize the performance benefits and not use the capacity multiplier function, while in your cost-sensitive application you may choose to max out the capacity multiplier to reduce the cost/GB of data stored.

All security features are built in. No need to buy additional licenses.

Absolutely. The drives are TCG Opal compliant and use AES-256 encryption to protect the data.

No, the use of the transparent compression function does not increase or decrease security. The drives meet TCG Opal standards for securing the data. https://ulinktech.com/tcg-certification-compliance-test-suite-list/

Security & reliability are aligned with enterprise SSD requirements (from OEMs, Hyperscalers, and industry standards). In terms of benefits, the CSDs can provide better economics ($/GB of data stored), better performance, better efficiency (performance/Watt, application performance per server), and better endurance (lifetime petabytes written) than other enterprise SSDs.

There aren’t really any new security vulnerabilities. The firmware is locked with no opportunity for users to modify it. The compression is done in a hardware engine, again not introducing any new attack vectors.

Nope! No new software or drivers to install, no application integration to do. If your application compresses data by default, you can change that setting to disable host-based compression, freeing up your CPU to work on the actual application processing.

The drives support any application, just as another NVMe SSD would. If the cloud part of the question refers to “apps hosted on a cloud service provider’s infrastructure,” the answer is still yes, the drives can work there. However, the big CSPs don’t usually tell you what specific storage drive your data or application is hosted on 😊For on-prem, the CSDs can be deployed for any type of application there are no limitations on the applications used with the ScaleFlux drives.

The only “integration” into a DB is really to turn off host compression if you have it on today.

Our WHQL certificate can be found here.

We have completed VMware certification and are working on VSAN. If you need VSAN support, please contact us to discuss it in detail.

We have completed VMware certification and are listed on their website. https://www.vmware.com/resources/compatibility/detail.php?deviceCategory=io&productid=57305&deviceCategory=io&details=1&partner=871&page=1&display_interval=10&sortColumn=Partner&sortOrder=Asc

Yes, the drives are hypervisor agnostic.

Yes! Customers report 2x or more performance.

No, there is not a dedicated video transcoding accelerator. Though this is a market we monitor closely.

The CSDs improve application performance in 2 ways: (1) the drives can respond with lower latency through higher workloads (as reported by Percona in their testing in a MySQL workload); (2) they reduce the burden on the CPU and the pollution of the host DRAM for the compression/decompression function, resulting in better system performance.

The engines and processing capability are aligned with the overall throughput of the CSD. Each CSD you add to your system adds just enough compute capability to handle the CSD’s tasks. Trying to use CPU cores to scale compression throughput to match the capability of multiple NVMe is a losing battle.

That’s one of the cool things with transparent compression – it works seamlessly with any application. The host sees a block storage device. The drive automatically compresses data on writes and decompresses the data on reads.

Across a variety of applications (databases, analytics, HPC, AI) customers report improvements in data access times (latency) and system responsiveness ( work per second). We’ve seen results up to 4x improvements.

Users will see the biggest benefits when they are using heavier workloads. In light workloads, the storage capacity gain will be the primary benefit. A Ferrari won’t get you there any faster if you don’t press the gas pedal 😊