Aerospike Summit 2020 spans May 12-14th and has a packed schedule with sessions by experts from around the industry. ScaleFlux will have a virtual booth and be mentioned in HPE's tech talk.
ScaleFlux® Runs Virtual Booth at Aerospike Summit 2020!
Aerospike Summit 2020 spans May 12-14th and has a packed schedule with sessions by experts from around the industry. ScaleFlux will have a virtual booth and be mentioned in HPE's tech talk. Below is a joint ScaleFlux-Aerospike blog. Check out our Summit companion page here!
Compression in ACT
Aerospike applications are very latency sensitive – particularly to outliers on Read Transactions. ACT is designed to assess how well SSDs manage read latency at various workload intensities. In order to “pass” at a given intensity, the SSD must have less than 5% of transactions exceed 1ms latency over a 24-hour test run. The higher the lower the percentage of read latency outliers at a given intensity, the better.
The most recent ACT 5.3 benchmark includes a data-compressibility parameter. While Aerospike deployments are primarily focused on performance and latency, users do also face budget constraints and need to save on their storage costs if possible. Much of the data used in an Aerospike database is compressible, with compression rates commonly 2:1 or higher. So, it makes sense to check into using a compression solution to take advantage of that compressibility for lowering storage costs... as long as there is no adverse impact on Aerospike transactions or latency.
The latest generation of ScaleFlux Computational Storage Drives, the CSD 2000 Series, leverages data compressibility to shine in the ACT test in three ways. First, the CSD 2000 passes the test at a whopping 525,000 transactions per second (over 200,000 TPS higher than the next-best 3.2TB NVMe SSD). Second, it delivers amazingly consistent latency across workload intensities, with 98% fewer read latency outliers than NVMe SSDs. Third, and most surprising, the CSD 2000’s integrated, hardware compression and variable-length mapping FTL, can simultaneously provide performance improvements and storage cost savings. For example, using the CSD 2000 at 2x its nominal capacity (i.e. setting the 4TB SKU to store 6.4TB of user data instead of the 3.2TB ordinarily utilized) cuts storage costs by up to 50% while still achieving higher TPS and a lower latency profile than ordinary NVMe SSDs. You can find this data in our Aerospike application brief: https://www.scaleflux.com/downloads/ScaleFlux_CSD2000_Aerospike_AB.pdf.
In customer testing, we have verified that a CSD 2000 with a raw, physical NAND capacity of 4TB can be used to store 6.4TB+ of user data. No other drive on the market has this capability.
“Hardware data compression in the drive enables users to push their Aerospike database to higher levels of performance and reduce their storage footprint,” said Paul Jensen, Vice President, Engineering Operations, Aerospike, “We see it as a natural extension to the high performance our customers expect from the world’s fastest NoSQL database.”
The ScaleFlux team is happy to have worked together with the Aerospike engineering team to bring compression parameters to ACT. If you would like to read more on the performance metrics of the CSD 2000 Series, please visit the ScaleFlux virtual booth at the soon-to-arrive Aerospike Summit 2020. We have also included more performance graphs and data in our companion site at https://www.scaleflux.com/aerospikesummit.html.
About the CSD 2000 Series
The CSD 2000 Series, available in industry-standard U.2 and PCIe Add-in Card form factors, integrates hardware compression/decompression into its SSD controller. The combination of this compression/decompression feature with ScaleFlux’s advanced Flash Translation Layer (FTL) is a unique offering in the industry. Unlike other compression options which introduce added latency, burden CPU cores and create system bottlenecks, the CSD 2000 enables users to take advantage of the storage cost savings of compression while actually reducing latency, improving transactional performance, and seamlessly integrating with mission critical database applications such as MySQL, PostgreSQL and Aerospike. Based on customer testing, the CSD 2000 typically stores 2-4x its physical capacity (e.g. the 4TB drive yielding 8-16TB of effective capacity!) – providing immense value compared to ordinary NVMe drives at a similar price point. Application performance also improves with the CSD 2000. Aerospike deployments, for example, see both a 50% uptick on ACT transactions per second and 2x effective storage capacity. MySQL and PostgreSQL testing shows similar gains, with additional applications in testing and qualification.
The ScaleFlux CSD 2000 Series is available now in 4TB and 8TB U.2, with the 4TB and 8TB Half-height, Half-length PCIe Add-in Card SKUs coming soon. To schedule a proof of concept, please contact your server vendor or ScaleFlux directly at [email protected]