Engineering Perspective: How the FAS3100 Nailed the Price/Performance Sweet Spot

When it comes to storage systems, low-end storage is focused primarily on achieving the lowest cost, while the high end is optimized to deliver performance and scalability. Between these two extremes is a growing midrange storage sector in which the goal is to optimize both characteristics and deliver the best price/performance.

This is what we were challenged to do with the new FAS3100 and V3100 series: optimize the platform to substantially increase performance for the midrange, while at the same time boost capacity and scalability.

Substantial improvements in performance and price/performance as demonstrated on leading benchmarks were driven by:

• New cost-effective single chassis and a two-power-supply HA-ready design
• A FAS3100 chassis design allowing focus on improvements to the heart of the storage system, the processor and memory subsystems
• Software optimizations

In this article we’ll detail the changes to the FAS3100 platform and discuss its performance on the SPC-1, SFS, and Exchange benchmarks.

What Is the FAS3100 Series?

The new series consists of two models: the FAS3140 (which is targeted to replace the FAS3020 and FAS3040) and the FAS3170 (which replaces the FAS3070). Equivalent V-Series systems provide similar benefits while front-ending legacy storage from IBM, Hitachi Data Systems, HP, EMC, Fujitsu, and 3PAR legacy storage platforms.

	FAS3020	FAS3040	FAS/V3140	FAS3070	FAS/V3170
Processors	2 × 32-bit (2 single-core)	4 × 64-bit (4 single-core)	2 × 64-bit (2 dual-core)	4 × 64-bit (4 dual-core)	4 × 64-bit (4 dual-core)
Memory	4GB	8GB	8GB	16GB	32GB
NVRAM	NV5 (1GB)	NV6 (1GB)	NV7 (1GB)	NV6 (1GB)	NV7 (4GB)
Expansion Slots	6 × PCI-X	6 × PCIe	8 × PCIe	6 × PCIe	8 × PCIe
Onbord I/O	8 × GbE 8 × 2Gb FC	8 × GbE 8 × 4Gb FC	4 × GbE 8 × 4Gb FC	8 × GbE 8 × 4Gb FC	4 × GbE 8 × 4Gb FC
Max Disk Drives	168	336	420	504	840
Max Capacity	84TB	336TB	420TB	504TB	840TB
Data ONTAP	7.0 +	7.2.1 +	7.2.5 +	7.2.1 +	7.2.5 +

Table 1) Comparison of FAS3100 series with FAS3000 series. (All numbers shown are for dual-controller, active-active configurations.)

Consolidated Chassis Design

The FAS3100 series chassis is a new, two-in-one design that houses two controllers in a single enclosure for active-active configurations. This approach was logical given that the vast majority of current FAS3000 series customers purchase their systems in this configuration. The chassis continues to provide redundant power and cooling, but requires fewer power supplies and other components, providing a more cost-effective approach versus the two discrete chassis of the previous generation.

Figure 1) The new FAS3100 series chassis versus the previous active-active configuration.

The new chassis offers several additional benefits:

• An integrated backplane reduces external cabling. Heartbeat communication (used in active-active configurations) between the two controllers is provided by an Infiniband connection built into the backplane. Two fewer power supplies also lead to a reduction of power cables.
• Single-controller configurations can be easily upgraded to active-active configurations by simply plugging in a second controller.

Processor, Memory, and Chipset Upgrades Boost Performance

The cost-effective system chassis design allowed us to focus on improvements in processors, I/O chipsets, memory, and NVRAM—the components that impact performance. The benefits of faster processors, the latest I/O chipset, and increased memory add up, yielding the significant performance boost we get from the FAS3100 versus the FAS3000 series, as described later.

Processor Selection
For midrange storage, we don’t choose the highest-frequency, bleeding-edge processors. Instead, we look for solid, proven processors that make economic sense. Processor vendors offer two road maps: one that includes the latest, fastest processors (where processor generations come and go quickly) and an embedded road map in which they commit to support individual processors for longer periods.

For the FAS3100 series we’ve chosen dual-core 64-bit processors from the embedded processor road map, enabling us to ship these systems without any changes to the processing subsystem for several years. This reduces our exposure to more frequent processor qualification and additional support complexity along the way. Enterprise customers standardize on a particular storage platform and want to purchase that platform for an extended period of time, so this strategy also helps us better serve these customers.

I/O Chipset
The FAS3100 series uses a newer generation HyperTransport™ chipset than the one originally introduced with the FAS6000 family. (NetApp’s use of HyperTransport chipsets was described in some detail in an earlier Tech OnTap article on the FAS6070) . While the newer HyperTransport chipset I/O characteristics are very similar to those of the previous generation, an increased number of buffer credits for PCI transactions in the current chipset provides a performance advantage for intensive I/O.

Memory and NVRAM
The final hardware investment we made for the FAS3100 series was to boost the amount of memory and NVRAM in the FAS3170. A dual-controller system now has 32GB of memory (16GB per controller) versus 16GB for the FAS3070. The memory we chose was also slightly faster, to further reduce memory bottlenecks. While both systems use DDR2 SDRAM, the FAS3070 used DDR2-400 while the FAS3170 uses DDR2-667 DIMMs.

NVRAM was also increased in the FAS3170 to 4GB in active-active controller configurations compared to 1GB in the FAS3070. This was done to maintain balance with the larger memory configuration. NetApp systems use NVRAM to temporarily cache writes to disk. Batching writes allows most data to be written in efficient stripes, including parity, rather than performing a much slower read-modify-write operation.

Optimized Software
With any hardware platform, as you add new features the software has to be optimized to take advantage of them. NetApp® software has now been optimized to take full advantage of the latest multicore chipsets; the FAS3100 series leverages these improvements with Data ONTAP® 7.3. The FAS3170 benchmarks covered in this article all used Data ONTAP 7.3 to maximize the resulting performance.

Enhanced Expandability
In terms of overall expandability versus that of the previous models, the FAS3140 supports an additional 84 disks over the FAS3040 for a maximum of 420 spindles (420TB using 1TB drives), while the FAS3170 supports an additional 336 disks over the FAS3070 for a maximum of 840 spindles (840TB).

We’ve also moved the NVRAM from a PCIe slot to the motherboard on both models. This makes an additional PCIe slot available—four slots in each controller for a total of eight expansion slots in active-active configurations. These slots can be used to support additional network ports or the new Performance Acceleration Module, an intelligent read cache that reduces latency and improves I/O throughput without additional fast disk drives.

FAS3100: Demonstrated Price/Performance

We’ve demonstrated the performance of the new FAS3100 platform using three different benchmarks:

• SPC-1, to assess SAN performance using an industry-standard benchmark that generates workloads typical of OLTP database activity
• SFS97, to assess NAS performance (NFS protocol)
• The Microsoft® Exchange Solution Reviewed Program (ESRP), to establish the number of Exchange users the storage system can accommodate

SPC-1 Price/Performance
A complete description of the SPC-1 benchmark was provided in a recent Tech OnTap article comparing the performance of the previous-generation FAS3040 relative to the EMC CX3 Model 40. In that testing, the FAS3040 achieved 30,985.90 SPC-1 IOPS™ at a cost per IOP of $13.61 versus $20.72 for EMC. (With snapshot technology enabled on both platforms the cost/IOP gap favored NetApp even more, with NetApp achieving $14.89 versus $54.49 for EMC.)

The new FAS3170 achieved 60,515.34 SPC-1 IOPS, nearly doubling the results posted on the FAS3040. Demonstrated price/performance was $10.01/SPC-1 IOPS, a substantial improvement over previously published performance numbers.

 

Figure 2) SPC-1 results for the FAS3040 versus the FAS3170.

Full SPC-1 results can be found at the following links:

• SPC-1 Benchmark Results (all)
• FAS3040 SPC-1 Results
• FAS3040 SPC-1 Results with Snapshot Enabled
• FAS3170 SPC-1 Results
• EMC CLARiiON CX3 Model 40
• EMC CLARiiON CX3 Model 40 with Snapshots Enabled

SFS97
FAS3100 performance was also assessed using the SPECsfs97_R1.v3 benchmark, which targets file service workloads by focusing on NFS version 3 throughput (operations/second) versus latency. The FAS3170 reached new heights here as well, demonstrating approximately 60% improvement over the previous-generation FAS3070, with a maximum of 137,306 ops/sec throughput. Latencies for all data points measured were also extremely low, resulting in an overall response time (ORT) of 0.94ms on the FAS3170 that peaked at just 2.1milliseconds for the final data point, compared to an ORT of 1.16ms on the FAS3070 with a peak of 2.9ms. SFS97 results for the FAS3070 versus the FAS3170.

Figure 3) SFS97 results for the FAS3070 versus the FAS3170.

Exchange Performance
Many customers use midrange storage systems to host Microsoft Exchange. The number of Exchange users supported by each storage system was assessed using the guidelines set out by the Microsoft ESRP. Using this assessment, the FAS3170 doubled the number of supported users versus the earlier FAS3070 model, scaling to over 50,000 Exchange users while using 188 disk drives to meet ESRP throughput and latency guidelines. For comparison, the high-end EMC Symmetrix DMX3 supports about 84,000 ESRP Exchange users but uses 448 disk drives to meet ESRP throughput and latency guidelines. From a price/performance perspective, we were very pleased with where the ESRP results for the FAS3170 landed. See full NetApp results.

Figure 4) Microsoft ESRP showing the number of supported Exchange users for each platform.

Conclusion

Optimizing the design of midrange storage systems is more subtle than designing high-end storage. The design implementation is done with customer needs in mind and therefore needs to be a careful balance between performance and price. The new NetApp FAS3100 series—along with its V3100 series companions for front-ending legacy storage—were designed to strike a balance between performance and price. This is born out by the impressive performance and price/performance of the platform as demonstrated on industry-standard benchmarks. The FAS3170 achieves 2x the SPC-1 performance of the FAS3040 plus a 25% improvement in price/performance ($10.01/SPC-1 IOPS versus $13.61/SPC-1 IOPS). The same platform also achieves 40% more SFS97 performance and supports 2x the Exchange users versus the FAS3070. Both systems are available and shipping now.

Mark Insley Senior Engineer, Platform Hardware NetApp Mark (left) is a key engineer in the midrange platform group responsible for the FAS3000 series. He was responsible for the electrical design of the FAS3020/FAS3050, the FAS3040/FAS3070, and the FAS3140/FAS3170. Prior to joining NetApp, Mark held a variety of hardware and software design positions, including work on the original Sun™ SPARCStation-1.

Lee Stork Senior Engineer, NetApp Lee's career in performance engineering began over 15 years ago at Tandem Computers, where Lee worked on processor, operating system, and database problems. Over the last three years, at NetApp, he has worked extensively on the FAS6080, the FAS3000, and the FAS3100 series of platforms. He is a graduate of the University of Nebraska, with a bachelor's degree in electrical engineering.