This Impact Brief by EMA highlights how caching in the server with FlashSoft 3.7 improves application performance, showing a 14x increase in IOPS Performance over HDD, and a decrease of storage latency of 22x.
Companies are looking to adapt quickly to market changes, meet customer demands, and capitalize on growth opportunities. To help meet these objectives, significant IT resources and effort is invested to maximize application performance. Higher application performance will influence how fast a transaction is performed, improve operational efficiency, and contribute to customer satisfaction. Servers and storage both play significant roles in application performance. Server performance has improved substantially over the last ten years, taking advantage of faster processors, increased memory, and optimized software. Although the improvement in server performance has been significant, storage performance has not kept pace. Hard disk drives used in storage systems have not changed much relative to performance, with little improvement in either latency or throughput. This creates a performance bottleneck at storage, leaving servers largely underutilized. This has limited any performance gains for applications. Applications can complete their tasks only as fast as data can be written to and read from storage.
Server virtualization has contributed further to the storage bottleneck problem. The aggregation of physical servers though virtualization has increased operational flexibility, lowered CAPEX costs, and reduced the environmental footprint and consumption of servers. However, fewer, more powerful physical servers increase storage workload densities, oversubscribing limited storage performance resources. The traditional approach of addressing storage performance issues by increasing the spindle count (adding additional hard disk drives) to increase performance cannot cost effectively scale to meet the performance potential of servers. The end result is that application performance does not meet its potential.
Flash technology is an effective means of addressing the storage bottlenecks, delivering very high input/output (I/O) storage performance even with highly random workloads. Flash solutions deliver significantly reduced response times, resulting in hundreds of thousands of input/output operations per second (IOPS). Flash is available in several forms: as server-side PCIe cards, SSDs in server-attached storage, all-flash storage arrays, and hybrid storage systems that utilize SSDs along with spinning disk drives. Of these options, server-side PCIe flash delivers the lowest latency and therefore the highest IOPS of any flash solution. This is because the server-side flash is physically adjacent to server processing, utilizing the server’s PCIe interface, which is much faster than a network connection. Networks require connectivity overhead that adds latency and decreases overall performance. An all-flash array that resides on a network specifies sub millisecond latency, where a PCIe flash solution has latency specified in microseconds.
PCIe flash located in the server is used to complement network-based SAN storage. The SAN storage, consisting of spinning disk drives, remains as the persistent, albeit slower primary storage. Caching, as we will see, is the functionality that enables these two storage types to deliver a cost effective yet high performance solution.
New servers are supporting more drive slots, providing an advantage where applications benefit with being co-located with storage . SSDs use an external hard disk package, but internally consist of flash chip technology. This enables SSDs to be used as a higher performing alternative in servers or storage arrays in place of hard drives. Using SSD within a server can scale flash to higher capacities that can be reached with PCIe server-side flash alone. Many environments now consist of both servers with direct-attached storage, as well as legacy SAN storage.
Since flash and SSDs cost considerably more than capacity spinning disk drives, the judicial use of these resources is recommended. Tier one, or the highest performing flash, is best limited to highly transactional workloads that can benefit from the higher performance that flash provides. There is a new lower cost, lower performing tier two flash that has been introduced in the market, and that use case will be the subject of a separate brief.
On March 31, 2015, SanDisk announced new capabilities of its FlashSoft software with version 3.7. FlashSoft is a member of SanDisk’s data center product family, which also includes PCIe application accelerators and solid state drives (SSDs). Changes to FlashSoft release 3.7 include performance, usability, and compatibility improvements for Windows®, Linux®, and VMware® environments.
In addition to the new release of FlashSoft, SanDisk is bundling FlashSoft 3.7 with its Fusion ioMemory PCIe application accelerators, the Optimus Ascend SAS SSDs, and the CloudSpeed Ascend SATA SSDs. The ioMemory3 is a server-side PCIe solution, and the Optimus and CloudSpeed SSDs are used within a server’s direct-attached storage. The chart to the right reflects the bundles that are available.
The SanDisk headquarters is located in Milpitas, CA. In addition to data center products, SanDisk also provides storage solutions for personal computing, mobile devices, and consumer electronics. More information on SanDisk can be found at http://www.sandisk.com
Flash is a having a significant impact on the enterprise, especially when it is deployed where it can impact performance the most: in the server. Flash can also be more economical when deployed in the server, as there is no incremental network-based storage array with additional hardware and software overhead that increases cost, power, and cooling. Deployment of flash is also easier within a server, as there is no need to migrate data to a separate device or set up incremental data protection and disaster recovery systems.
The amount of flash that can be deployed within a server is limited and remains much lower in capacity than the primary storage. The key in getting the most benefit from server flash lies in software that maximizes the efficiency of server flash. This increase in efficiency is where SanDisk’s FlashSoft software excels, as it provides caching of data within the flash. Caching is the process of keeping the most active data in flash, so that the hotter data, also referred to as the “working set,” can be accessed much quicker than if it was accessed from primary storage. Since the amount of flash capacity is much less than primary storage, having an effective caching algorithm is the key to maximizing high performance. Having the correct data in cache reduces the chance for read misses, increasing overall read performance. FlashSoft, with its sophisticated caching algorithm, makes the correct caching decisions based on analyzing workload trends along with identifying dynamic working-set characteristics.
FlashSoft supports two forms of caching, write-through or write-back methods. Write through-caching benefits workloads such as web servers, search and discovery, analytics, and streaming media Write back caching optimizes performance with workloads such as databases and transactional applications.
Write-through caching writes concurrently to both cache and primary disk at the same time. Although there is no performance benefit to the write operation, subsequent reads will be already resident in cache. This takes advantage of a well-founded assumption that data that has recently been written is the data that will be most likely be read shortly thereafter. By having the data already in cache, the initial time reading data from slower primary storage is not required.
FlashSoft also utilizes write-back caching, an approach that provides the greatest performance benefit to write intensive workloads. By storing data initially to cache and then writing data after the fact to primary storage, write operations benefit immediately from a performance boost. FlashSoft addresses error handling requirements in the event that there is a failure in flash before the storing of data in primary storage through the use of either hardware- or software-based RAID.
There is a situation where writing all data into cache first is not effective. When large contiguous write operations are performed, the cache can be tied up with a phenomenon known as “cache bottleneck” that slows down overall system performance. FlashSoft uses a feature called Write Stream Recognition that recognizes the contiguous writes, and bypasses cache writing to go directly to primary storage. The threshold at which the corrective action is determined can be configurable within the FlashSoft software.
FlashSoft has a number of other benefits that differentiates it from alternative flash software. These benefits include the following:
There are a number of new capabilities with FlashSoft 3.7 that increase the usability and impact of the software and flash hardware. The following capabilities per platform are now available.
When flash is being considered as way to accelerate applications, cost can be an immediate concern. However, there are a number of benefits to flash that should be considered that will actually lower overall costs.
Flash hardware and software solutions for servers can cost under $10,000 and still deliver a huge boost in application performance. This is significantly less than what is required to deploy an all-flash or hybrid array. The performance in IOPS for data within flash vs. hard disk drives is significant and well documented, but the performance benefit can also be significant when characterized across the entire storage infrastructure. Performance tests illustrate FlashSoft software was able to accelerate a workload from 450 IOPS to 6462 IOPS in a Windows environment—over a 14x increase in performance compared to the non-accelerated back end. Storage latency was also greatly improved, dropping from 1272 milliseconds to 55.65 milliseconds—over 22x improvement in response time. Applications are measured by how much work they can get done in a specific amount of time. By increasing the amount of work by 14x with a nominal investment, then the overall cost per application transaction is greatly reduced.
By minimizing latency with server flash, the amount of I/O between server and primary SAN storage can be greatly reduced. This is because network-based storage is not only provisioned for capacity, but for performance as well. Additional spindles, HBAs network switches, and storage controllers that are normally deployed to increase performance can be avoided. It is not uncommon that a solution that utilizes FlashSoft with flash can save up to 25% or more in storage costs compared with the same number of servers that utilized direct-attached and/or SAN storage alone.
FlashSoft software supports not only SanDisk PCIe flash and SSDs, but those of third party hardware vendors as well. This capability was developed before SanDisk acquired FlashSoft in 2012. SanDisk was smart to keep the multiple hardware vendor capability in place. This capability provides yet more investment protection for customers. In addition to supporting pre-existing SanDisk PCIe and SSD customers, SanDisk is able to gain new customers for both its hardware and software.
The improved FlashSoft software and solution bundles are welcome additions to the all-flash market, and EMA applauds SanDisk for making flash and SSD solutions for the enterprise more productive, cost effective, and easier to deploy.
Whether you'd like to ask a few initial questions or are ready to discuss a SanDisk solution tailored to your organizations's needs, the SanDisk sales team is standing by to help.
We're happy to answer your questions, so please fill out the form below so we can get started. If you need to talk to the sales team immediately, please phone: 800.578.6007
Thank you. We have received your request.