Multi-Core CPUs Turbocharge All-Flash Storage Systems

by Larry Freeman, NetApp

Free Download

IDC White Paper: NetApp All-Flash Storage for the Enterprise

Learn More

Server vendors long ago transitioned to multi-core CPU architectures to improve price-performance and maximize the number of virtual servers that can run on a single physical machine. More recently, the I/O demands of fast servers and solid-state storage devices have required storage vendors to make a similar transition. As was the case with servers, the transition requires storage vendors to design their operating systems (OS) to take advantage of a multi-core architecture.

This article examines two all-flash storage platforms—NetApp EF-Series and NetApp FAS systems—and how the storage OS for each uses multi-core CPUs to drive high performance.

SANtricity: High IOPS at the Lowest Latency
EF-Series systems run the SANtricity OS, a real-time operating system built on a lightweight kernel designed solely for high-performance and low-latency storage operations. With an efficient instruction path length measured in the 30-microsecond range, SANtricity imposes minimal constraints on the processing of I/O instructions.

To take advantage of multi-core CPU architectures, such as the Intel Haswell Xeon processors used in the EF560 all-flash array, SANtricity was enhanced to include a scalable I/O scheduling method that uses multiple cores to execute code segments in parallel, with individual tasks selectively assigned to separate execution engines (or cores). However, multi-processing operations are executed only when storage I/O performance will benefit from running on multiple cores. This opportunistic use of multi-core processing keeps the I/O path short and enables the storage controller to sustain extremely high IOPS at the lowest possible latencies.

Flash Storage For Dummies, NetApp Special Edition: Optimize performance and reduce the footprint of storage infrastructure in the data center.

EF-Series Multi-Core Performance
A NetApp EF560 all-flash system was tested using the SPC-1 storage performance benchmark, which simulates random block-access workloads, such as OLTP systems, database systems and mail servers. In the SPC-1 tests, the EF560 posted one of the best combinations of all-flash price and performance for systems with an average response time of less than 1 millisecond (in storage performance testing, the terms ‘response time’ and ‘latency’ are often used interchangeably when referring to the time it takes for a storage system to complete an I/O request).

EF560 performance highlights include:

  • 650,000 sustained IOPS with response times in the 800 microsecond range1

  • 245,011 SPC-1 IOPS with an average response time of 930 microseconds

  • 122,491 SPC-1 IOPS with an average response time of 340 microseconds

SPC-1 Benchmark Reports

NetApp EF560

Executive Summary (PDF)

Full Disclosure (PDF)

Data ONTAP: Scalable IOPS, Low Latency and Advanced Data Management
NetApp FAS systems run the Data ONTAP OS, which IDC consistently ranks as the number one branded storage OS within the open networked disk storage systems market based on revenue shipments. Data ONTAP enables storage clusters to scale performance by adding all-flash FAS nodes, and it provides a wide range of advanced data management features, including snapshots, replication, compression, deduplication and data protection.

For Data ONTAP, the secret to maintaining high performance while handling a wide range of data management tasks is to fully utilize every available CPU core. To accomplish this, processes are scheduled across all cores to reduce latency in the CPU complex and enable simultaneous support for both front-end workloads and back-end data management. Performance testing using a database workload has validated the ability of Data ONTAP to simultaneously utilize every available core at close to maximum load (over 95% utilization per core).

All-Flash FAS Multi-Core Performance
Like the EF560, Fibre Channel-attached, all-flash FAS8080 clusters deliver impressive results for OLTP workloads, with an 8-node cluster rated up to 1,440,000 sustained IOPS with response times under 1 millisecond2. FAS clusters scale performance linearly as HA pairs are added, and all-flash nodes can be added nondisruptively to keep pace with new performance demands. In addition, workloads can be moved and replicated between all-flash and hybrid-flash nodes within a cluster to achieve the best balance of price and performance for specific use cases, such as desktop virtualization, OLTP and disaster recovery.

Flash-Optimized and Highly Reliable
In addition to delivering high performance, NetApp EF-Series and FAS systems have both been validated for over five nines of uptime across a large population of systems. This means that your mission critical applications can benefit from flash-optimized, multi-core performance and still experience the high reliability required to keep your business running smoothly.

1 Based on NetApp internal testing using 100% 4K read operations
2 Based on NetApp internal testing using 100% 8K read operations

No portions of this document may be reproduced without prior written consent of NetApp, Inc. Specifications are subject to change without notice. NetApp, the NetApp logo and Go further, faster, are trademarks or registered trademarks of NetApp, Inc. in the United States and/or other countries. Click here for a full listing of NetApp trademarks. All other brands or products are trademarks or registered trademarks of their respective holders and should be treated as such.

Show more