White Paper Extending NetApp Deployments with stec Solid-State Drives and Caching
Contents Introduction Can Your Storage Throughput Scale to Meet Business Demands? Maximize Existing NetApp Storage Investments with Caching Automated Cache Configuration Summary 3 4 5 6 7
In today s typical high performance business environment, the number of users concurrently accessing applications places a significant performance demand on storage subsystems competing for resources on virtual machines. This translates into more expensive storage for the increased Input-Output Operations per Second (IOPS) required to deliver application performance. stec resolves these issues by helping customers maximize performance with a cost-effective solution combining stec s solidstate disk drive (SSD) technology and stec s EnhanceIO SSD Cache Software. 03
1 Can Your Storage Throughput Scale to Meet Business Demands? Historically, as application demand grows, additional virtual machines (VM) or ESX hosts are installed. As these hosts increase the IOPS demand on the Filer storage subsystem, additional drive trays are added to scale up to the desired throughput rate. The NetApp controller hardware architecture has a maximum throughput rate of 2GB/s, and Romley-class servers can sustain a 5GB/s rate, so the storage subsystem becomes a bottleneck despite adding more drive trays (refer to Figure 1). IOPS performance scaling is frequently calculated on a per-hard disk drive (HDD) basis of 150-200 IOPS. If 10,000 IOPS are required, a minimum of 50 HDDs are needed. This scaling model rapidly breaks down when cost, space, power and the continuous requirement for greater throughput are evaluated. A recent ESG report highlights more than 2X growth in VMs per ESX host from 10 (2008) to 24 (2012), and it s expected to continue at the same rate over the next four years, thereby compounding the mismatch between the data rate a host server can request and what a Filer can deliver. Figure 1: ESX Host Server and NetApp Storage Array App ¹ A ¹ ESX HOST STORAGE ARRAY FAS 3200 (2GB/s) App 2 App 3 Disk Shelf 1 Disk Shelf N App N Disk Shelf 2 This performance gap is further exacerbated when the read/write mix through the ESX host is randomized. The virtualization blender effect transforms application workloads even if they are generally sequential into random workloads. The blended or random workload heavily taxes the storage subsystem because HDDs incur milliseconds of latency to seek to the proper location on the disk. As the I/O throughput climbs toward the Filer s 2GB/s controller bandwidth limit, write commands will be prioritized over reads to ensure data coherency. This in turn reduces the available throughput for read operations degrading throughput for all applications and users. 04
2 Maximize Existing NetApp Storage Investments with Caching Software A simple, effective and inexpensive solution to extending the useful life of an existing NetApp investment is to add SSD read caching to the NetApp storage architecture. There are two components required: an SSD, and a caching software application that resides in the host. See Figure 2. App ¹ A ¹ ESX HOST s1120 PCIe STORAGE ARRAY FAS 3200 (2GB/s) Figure 2: ESX Host with PCIe SSD Host Adapter Configured as Cache App 2 App 3 Disk Shelf 1...Disk Shelf N App N Disk Shelf 2 The cache illustrated in Figure 2 is implemented with PCIe SSD technology. SSDs are capable of 40,000 to 100,000 random IOPS, making them ideally suited to solve the highly randomized workload generated by the ESX host. Complementing the SSDs, a software caching application in the Guest VM or ESX host provides the intelligence to automatically migrate copies of frequently accessed data to the SSD from the Filer s HDD storage. stec s EnhanceIO SSD Cache Software delivers the requested data to the application from the low latency, high throughput SSD, not from the Filer s slower HDDs. This immediately reduces the number of read transactions the Filer must process. By offloading read operations to the SSD, more bandwidth is now available in the Filer to process write commands. Greater throughput for both read and write commands is achieved, and the useful life of the Filer can be extended. This strategy does not change where the authoritative data is stored, so existing investments in NetApp s valuable utilities (backup, restore, mirroring, etc.) are maintained. 05
3 Automated Cache Configuration Implementing the proper cache parameters is a function of the read/ write workload issued to the storage subsystem. The objective is to keep the most frequently accessed data in the cache, where it can be delivered to the application as quickly as possible. The EnhanceIO caching solution includes a Profiler utility that automates the analysis of the read/ write workload, calculates the recommended cache size, and includes a measure of how much performance can be improved by implementing the recommended cache parameters. The monitor only mode of the Profiler can be installed and run prior to adding SSDs to the system. Once the Profiler operation is completed, the results are displayed for the administrator, who can then make a pointand-click implementation of the proper cache parameters. This helps administrators determine exactly how much cache is required. Figure 3: ESX Host and Storage Array with Multiple Cache Implementation Options App ¹ A ¹ ESX HOST s1120 PCIe STORAGE ARRAY FAS 3200 (2GB/s) App 2 App 3 Disk Shelf 1 Disk Shelf N Disk Shelf 2 App N ESX HOST w/ stec s800 Series SAS SSDs OR SAS SSD JBOD /w stec s800 Series SAS SSDs 06
4 Summary The SSD-based caching approach increases the available bandwidth in the NetApp storage array, increases the overall system throughput, and does not require significant investment in more storage. This approach does not impact NetApp s suite of management utilities, so disaster recovery, availability and manageability remain unchanged. Multiple VMs can be individually enabled for caching as required. Two of the most time-consuming aspects of cache configuration are determining how much cache is required, and what files should be cached. Answers to these questions are automatically determined by the EnhanceIO Profiler application, making initial configuration and ongoing tuning efforts very simple to administer. The EnhanceIO application code base and management utility is consistent across Windows, Linux and VMware, further enhancing ease of use. Multiple implementation options provide different architectural approaches. Figure 3 illustrates the three options where the SSD cache can be deployed. The highest performance option is usually achieved by installing an s1100 Series PCIe card in the ESX host. Alternative implementations include using one or more s800 Series SAS SSDs in the ESX host or in an external JBOD enclosure. About stec stec, Inc. is a leading provider of enterprise-class, solid-state storage solutions designed for the ever-growing performance, reliability and endurance requirements of today s advanced data centers. The industry s first company to deploy solidstate drives (SSDs) into large-scale enterprise environments, stec offers the industry s widest range of solid-state storage solutions, which protect critical information for major business and government organizations worldwide. Headquartered in Santa Ana, California, stec also serves the embedded and military/aerospace markets with SSDs for industrial and rugged environments. Enabling Tomorrow s Applications For more information on stec products, solutions and technology, please visit www.stec-inc.com facebook.com/userstecinc twitter.com/stec_inc youtube.com/user/stecincssd 3001 Daimler Street, Santa Ana, CA 92705 +1.949.476.1180 2013 stec, Inc. The stec name, logo and design are trademarks of stec, Inc. All other trademarks are the property of their respective owners. 1305_50V01_US