Data Center Tech Blog

Abandoning the decades-old SAS and SATA software protocol stacks, NVMe is the new kid on the block. Will it dominate the data warehouse workload space?

Peter Plamondon

Director, Strategic Partner Ecosystem

If your favorite sports team were the champion in 9 of the last 16 seasons, a 56% championship percentage, that would be pretty impressive. What would you call another team that soundly beat that nine-time champion – perhaps a “champion in the making”?

When it comes to data warehouse workloads, the championship that every server OEM seeks to win is the TPC-H world record benchmarks. With the introduction of NVMe, will it dominate the data warehouse workload space?

Let’s Talk about the TPC-H Nine-time Champion

For the last two and a half years, leading server vendors Lenovo™, Cisco®, and Hewlett Packard Enterprise selected the SanDisk® Fusion ioMemory™ PCIe Application Accelerators to power 9 of their TPC Benchmark™H (TPC-H) ) non-clustered world records.

TPC-H is a well-known, industry-standard, decision-support benchmark used since 1999 to illustrate decision-support systems that examine large volumes of data, execute queries with a high degree of complexity, and give answers to critical business questions. TPC-H and other benchmarks were created by TPC, a non-profit corporation, to define transaction processing and database benchmarks and to disseminate objective, verifiable TPC performance data to the industry. You’ll see the workload used for the TPC-H benchmark referred to as a data warehouse, business intelligence, or data analytics workload.[1]

The Fusion ioMemory devices use flash memory and the PCIe bus to deliver high read bandwidth of 2.8GB/s, and up to 6.4TB[2] of capacity. For performance-sensitive workloads like data warehouse and TPC-H, that capacity and performance is just what a server needs to deliver a great TPC-H benchmark result.

It’s not surprising that leading OEMs used 4, 8, or even 16 Fusion ioMemory devices to power their TPC-H benchmarks.

NVMe and Data Warehouse Workloads – the New Kid on the Block

 Time marches on, new technologies appear, and customers’ needs change.

The new kid on the block is NVM Express® or NVMe™, an industry standard protocol designed from the ground up to fully exploit the capabilities of non-volatile memory technologies (i.e., flash). NVMe delivers high bandwidth and low latency, in part, by abandoning the decades-old SAS and SATA software protocol stacks created for much slower hard disk drives – that’s the “from the ground up” part, no legacy protocol stack.

NVMe brings the PCIe bus signals over a ribbon cable and a SAS-like connecter to 2.5” U.2 Small Form Factor devices. NVMe devices are also available in the traditional Add-In Card (AIC) form factor like HH-HL (Half Height, Half Length).

With PCIe performance now available from NVMe U.2 devices, OEMs can maintain system performance while reducing the number of PCIe slots, making more efficient use of limited space within the server. Many NVMe U.2 devices – including the Ultrastar® SN200 – support hot swap, which eliminates the need to power down a server to swap an add-in card.

NVMe also benefits from the increased bit density in NAND flash, which enables higher capacities in less space. For example, the 6.4TB Fusion ioMemory SX350 AIC is a Full-Height, Half-Length (FH-HL) card, compared to the 6.4TB Ultrastar SN260 AIC which is a Half-Height, Half-Length (HH-HL) card. The Ultrastar SN200 U.2 SFF device is also available with 6.4TB capacity.

An important note for high performance is that the AIC form factor uses 8 PCIe lanes compared to 4 lanes used by the U.2 form factor. Servers focused on maximum performance, like a 4- or 8-socket server, continue to provide several PCIe slots to support maximum storage performance.

The server ecosystem has embraced NVMe for all these great reasons. Modern operating systems include NVMe device drivers, and server chassis that accommodate 20, 24 or 48 NVMe devices are available.

NVMe and Data Warehouse Workloads – How Does it Compare to ioMemory?

We have recently introduced the second-generation HGST Ultrastar® SN200 Series NVMe drives in capacities from 400GB to 7.68TB and with exceptional I/O performance and ultra-low latency.

Let’s compare the key performance metrics for data warehouse workloads from the datasheets of HGST Ultrastar SN260 AIC, HGST Ultrastar SN200 U.2, and Fusion ioMemory SX350 AIC products.

NVMe data warehouse speed benchmark

[4] [3]

Sequential Read Bandwidth is key for data warehouse workloads, and the first row of the table above highlights the benefit of the Ultrastar SN260 AIC device using 8 PCIe lanes to deliver 6.17 GB/s bandwidth, compared to the Ultrastar SN200 U.2 device using 4 PCIe lanes to deliver 3.35 GB/s.

Data Warehouse Test Results Using HGST Ultrastar SN200 Series

If you’re attending Microsoft® Ignite, come to HGST booth #857 to find out if a server using the Ultrastar SN200 Series dominated the earlier results that used Fusion ioMemory. 🙂

Otherwise, watch this blog for a follow-up post with benchmark results. You won’t have to wait long.

Is NVMe the New Champion?

When it comes to NVMe and data warehouse workloads, we’ll have to wait for publication of well-known, independently validated industry benchmarks before we can officially crown a new champion. But it might be time to bring the crown out of storage and polish it up, just in case.

How can you best leverage NVMe for your workloads and within your infrastructure? Stop by the HGST booth #857 at Ignite to learn more about our products, data warehouse benchmark results, and how you can take advantage of one of the industry’s largest storage and systems portfolio. Our experts will be happy to share our deep learnings with you.

Learn more about how HGST will light up Ignite 2017.

[1] TPC Benchmark™, TPC-H, QppH, QthH, and QphH are trademarks of the Transaction Processing Performance Council.

[2] One terabyte (TB) is equal to 1,000GB (one trillion bytes) when referring to solid-state capacity. Accessible capacity will vary from the stated capacity due to formatting and partitioning of the drive, the computer’s operating system, and other factors

[3] 128KiB block size.

[4] 4KiB block size.

7 Amazing Facts About the Big Data of Blockbuster Animations

The Dynamic Value of Storage?

bring your data to life

Today’s digital economy with mobile, IoT and cloud is based on the value of data. How do you unlock it? Download the infographic to learn more: