Flash Memory Arrays Enabling the Virtualized Data Center. July 2010

Transcription

1 Flash Memory Arrays Enabling the Virtualized Data Center July 2010

2 2 Flash Memory Arrays Enabling the Virtualized Data Center This White Paper describes a new product category, the flash Memory Array, and how it meshes seamlessly into the current evolution of data centers and cloud computing facilities. Violin Memory pioneered and leads the Memory Array market segment allowing the Enterprise to realize the long-held ideal of balance among compute, storage, and networking resources. Violin s 3200 product line brings flash memory into the data center with inherent reliability (flash RAID), sustained performance, spike free latency, and Enterprise class availability, reliability, and serviceability. Violin s switched memory architecture allows for uniquely scalable and reliable aggregation of memory of any type into a sharable Memory Array. Violin s Memory Arrays offer better price and performance metrics than traditional performance storage with or without SSDs. Enterprise customers now have a true choice when faced with their next performance storage purchase 10x the performance in 1/10 the space or to continue with their legacy HDD solutions. Virtualization continues to spread throughout the data center at an unrelenting pace. Server virtualization coupled with multi-core processors means as many as 20 virtual servers can be hosted on one physical platform. Storage virtualization reduces the number of storage systems and the ultimate capacity required. And in the network, virtualization promises to bring about a convergence with storage protocols which will eliminate much of the cabling and drastically reduce the number of switches and adapters required. One element continues to act as the bottleneck spinning hard disk drives (HDDs). While the latest multi-core CPU s process billions of operations a second, and local memory can deal with tens of millions of operations a second, hard drives move at a far more leisurely pace. The fastest disks around have access times averaging about 5 milliseconds, which equates to 200 operations per second. With RAID, disk array architectures and memory caching certainly improve the rate of data transfer but there has been no getting around the fact that spinning disks are the weak link in overall system performance.

3 3 In the last couple of years, flash technology has been introduced as a solution to the limitations imposed by the disk bottleneck. It has been very successful in laptops where portability is valuable and in high performance computing installations where price is no object. Flash improves performance immensely compared to hard drives particularly when it comes to read Input/Output Operations per Second (IOPS). But to be a true enterprise-ready solution, several noteworthy challenges must be resolved: Writes are sequential and relatively slow (200 1,000 µsec). Erases take considerable time (1,300 to 10,000 µsec). During this time, nothing can be read or written. Flash memory wears out after a certain number of Erases The cost of flash is increased by the need to overprovision for performance Efficient RAID algorithms like RAID 5 and RAID 6 cannot be used with Flash and instead inefficient RAID 1 (mirroring) is commonly utilized. Low latency hot-swappable interconnects (suitable for Flash) are required DRAM solves many of the problems of traditional flash, however it is not persistent and has a much higher cost per GByte. In most cases, it just doesn t make economic sense to load up servers with a vast battery of memory. But for flash to become pervasive in enterprise data center applications, it requires a very different set of attributes than those provided by SSDs. Fortunately, the issues can be overcome through proper design of large-scale flash arrays. Violin Memory s architecture enables thousands of flash devices to operate efficiently as a Flash Memory Array : Writes are random and fast (20 µsec) Erases are simple, fast and hidden from the user no multi-millisecond latency spikes Reading is accelerated through parallel access to thousands of flash chips Data integrity is maintained with efficient flash RAID protection for reliability and lower cost High performance flash controllers and hardware flash RAID reduce the latency of the flash storage Maintenance and serviceability (hot-swap & fail-in-place) are integrated into the architecture Violin s architecture is based on a high-performance flash controller and a purpose-built and patent-pending flash RAID algorithm. This results in very high sustained performance for Writes as well as Reads. A single shelf Memory Array delivers the IOPS of a thousand HDDs with 95% less latency.

4 4 User Benefits: 10x Performance Increase By taking advantage of a Violin Memory Array, applications and tasks typically run ten times faster. Transaction/query times and batch analytics benefit enormously; jobs that run in 2 hours now finish in 10 minutes. Customer and user satisfaction are significantly increased by heightened system responsiveness. Take the case of database analytics. System Engineers no longer have to waste time squeezing out the next 2 percent performance gain by changing the database layout. Instead, they are free to focus on applications that generate real value for the organization. Desktop applications, web serving applications and VMware productivity can be increased through shared access to high performance storage. A Violin Memory Array connected via a 1Gbit/s Ethernet connection (220K IOPS) is significantly faster than the local disk (<200 IOPS). The real question for the enterprise data center administrators is do I really need to buy another SAN array? Financial Benefits: Significant Cost and Power Savings A Violin Memory Array packages higher performance with significant cost savings. It facilitates application acceleration and server consolidation while significantly reducing the total power requirements. Let s look at the facts: A. HDDs spin continuously and large numbers are needed to meet the aggregate demand of enterprise applications where I/O is the typical bottleneck. A typical high-end HDD supports 144GB of data, 300 IOPS and consumes 20W of power. Adding RAID protection overheads, over 1,000 drives and 20kW are required to support I/O rates of 220,000 IOPS and transfers of 1 GByte/sec. Violin achieves this performance with just 20 flash memory modules (VIMMs) consuming less than 300W of power. B. Servers and CPUs in the data center are usually poorly utilized. CPUs consume 200W on average, yet operate at less than 5% utilization. Virtualization technology is useful in consolidating many applications onto a single server. However, I/O and/or storage performance limitations force organizations to deploy massive local memory footprints for each application. Violin resolves these bottlenecks by reducing latency from 5ms to 200 microseconds, a 25x improvement which significantly increases CPU utilization. This dramatically lowers the need for server memory and CPUs while enabling more efficient virtualization and server consolidation.

5 5 C. PCIe flash cards are faster than SSDs, but are tied to individual servers. This makes sharing the data, adding capacity and servicing the servers much more complex as the servers must be taken off-line and any performance gained cannot be shared. Adding flash capacity to existing servers is significantly more complex. D. Legacy storage systems featuring SSDs are designed to integrate flash memory within existing HDD enclosures and use software-based RAID controllers. This retrofitting approach results in (delivered) costs of approximately $200 per GByte (RAID-1) and performance in the 10-30K IOPS range with latency typically around 2ms or more. Violin Memory Arrays, on the other hand, can easily integrate into the existing storage network. Costs are now below $20 per GByte and price/performance effectively increases by 500% or more. Strategic Benefits Organizational productivity and agility are directly tied to IT systems and their data center architectures. Ideally, organizations can rapidly implement new applications and mine existing data for important information. However this is seldom the case when HDD storage is tuned to operate for specific workloads, applications and/or databases. Introduction of a new application can negatively impact data center performance and require radical, resource intensive, architectural changes. Violin s Memory Array adds significant flexibility and headroom to application performance. System workloads can be increased with little or no change in architecture. Like a high performance engine, when you push the accelerator, speed is not constrained. With this approach, moving from 5K transactions per second to 220K requires no additional hardware or software other than the introduction of a flash Memory Array. In many cases a Violin Memory Array might allow the organization to substitute for a more expensive hardware upgrade project that would entail the addition of racks of hard disk drives and/or bringing in additional servers and power. Violin 3200 Memory Array The Violin 3000 Series is the latest generation of Violin s Memory Arrays. The first of this series, the Violin 3200 provides over 10TB of flash capacity with integrated flash RAID and 250K IOPS. Violin s platform is purpose-built to enable cost-effective large-scale deployment of flash in the enterprise data center.

6 6 By augmenting existing HDD storage with a Violin Memory Array, the primary benefits are: 50x reduction in latency 20x reduction in cost per IOP 50x increase in IOPS per shelf The Violin 3200 enables the virtualized data center by removing the cost and performance barriers to high-performance virtualized storage. It is ideal for applications with large amounts of data, which require rapid access and active processing. This includes: Databases Analytics Data warehousing Messaging systems Document control Storage metadata caching and tiering The Violin 3200 delivers the industry s highest performance and capacity per dollar as a shared RAID-protected storage system. For instance, 14 Violin 3200s in a single rack deliver 3.5 Million IOPS and 14 GB/s throughput. Violin s architecture is optimized to deliver the industry s lowest spike free latency typically 95% lower than HDDs and 70% lower than OEM integrated SSDs or PCIe cards under typical Read/Write loads. Violin s Arrays are designed for % system availability. This is achieved by using a patented RAID algorithm with spare flash devices and spare hotswappable modules. Modules can fail-in-place with up to 4 spare modules available in a single platform. For more information see Violin s Flash RAID White Paper available from the Violin Memory website. Violin s Memory Array brings the balance of compute, networking, and storage as independent integrated nodes of functionality. The combination of server and storage virtualization in tandem with Violin Memory arrays sets the stage for a truly virtualized data center without application specific silos. That adds up to doing much more with much less.

7 7 Contact Violin Violin Memory, Inc. USA 2700 Garcia Ave, Suite 100, Mountain View, CA ) 9- VIOLIN Ext 10 or (888) Ext 10 memory.com memory.com