WHITE PAPER Accelerate the Performance of Virtualized Databases Using PernixData FVP Software Increase SQL Transactions and Minimize Latency with a Flash Hypervisor 1
Virtualization saves substantial time and money by reducing hardware and software infrastructure costs and improving IT operational efficiencies. However, as more applications become virtualized and Virtual Machine (VM) densities increase, the storage array can t keep up with the load. This I/O contention at the storage layer has caused many companies to experience a drop in performance for virtualized databases (e.g. Oracle, Microsoft SQL Server, MySQL). This is because these databases are especially susceptible to I/O bottlenecks in the storage array, where slow reads and writes can make it difficult to run reports and complete transactions expeditiously. PernixData FVP solves database performance problems easily and cost effectively. It is a 100% software solution that accelerates reads and writes to primary storage by clustering server-side Flash. With FVP in place, database response times are reduced by 60%, more than doubling the number of orders that can be performed per second (a 129% improvement). This makes PernixData FVP a strategic part of any virtualized database deployment. Challenges with Virtualized Databases Generally, databases fall into two categories, Online Transaction Processing (OLTP) and Online Analytical Processing (OLAP). OLTP databases require low latency storage in order to execute many transactions in a short timeframe. OLAP databases require high I/O throughput to process large volumes of information in a timely manner, typically for tasks like reporting. In a physical (non-virtualized) environment, Database Administrators work with IT to optimize the infrastructure for maximum database performance. This typically involves setting up dedicated infrastructure of compute, network and storage resources for database activities. The storage infrastructure must also be specifically tuned for the type of data (random vs. sequential) and operation (read vs. write). CONTENTS: Challenges with Virtualized Databases 2 Flash to the Rescue 2 Flash Hypervisor Acceleration 3 Real World Examples 4 Summary 5 In a virtual environment, however, the infrastructure is shared across a number of applications. As many VMs vie for the same I/O resources, contention is inevitable as the hypervisor tries to schedule the available resources fairly across all the VMs. Without a good way to optimize performance for individual workloads, it is extremely difficult to guarantee the low latency and high throughput needed for virtualized databases. This can cause undesirable consequences, including: Inability to virtualize all databases. In some cases, virtual databases must be switched back to physical servers. Over-provisioning of storage infrastructure to support database requirements. Poor performance of virtualized databases. Flash to the Rescue Many companies are looking at Flash storage as a solution for poor database performance in virtual environments. One option is to put Flash in the storage array to lower the latency associated with reads and writes to primary storage. This may improve application performance, but it is quite expensive and does not solve all database performance issues. There may be additional I/O bottlenecks because all VMs accessing data on a storage array must traverse the storage fabric and the storage processor before ever reaching the storage media (in this case Flash). So, even if Flash in the storage array increases the speed of reads from and writes to media, bottlenecks in other areas still exist, causing suboptimal database performance. 2
A better approach is to use Flash in the host where the database resides. Server Flash can then accelerate reads, avoiding many of the bottlenecks described above. This results in extremely low SQL latency at a fraction of the cost of Flash in the storage array. Server-side Flash has its own challenges, however. These include: Requires changes to VMs. Does not fully support virtualization cluster services. Typically only optimizes read-intensive workloads. Flash Hypervisor Acceleration PernixData FVP solves the above challenges and delivers the best possible performance for virtualized databases in the simplest and most cost-effective manner. FVP is Flash Hypervisor software that clusters server Flash into a logical pool of resources used to accelerate reads and writes. It can optimize both read- and write-intensive workloads, making it ideal for virtualized applications. FVP installs inside the hypervisor in minutes, and requires no changes to the database, other applications, or the storage array. Once in place, FVP delivers faster database response times, which results in more transactions in any given time period. In internal testing, for example, the DVD Store application was used to generate load against a Microsoft SQL Server instance deployed in a VM. The DVD Store application measures the orders processed per second as well as the time to complete an order. Note that a single order is composed of multiple transactions, such as customer logins, browsing for different titles, and placing orders. In addition, each transaction is composed of multiple read/write SQL statements. The key metric in this test is Orders per Second. With the standard database, 75 orders are processed per second while the FVP accelerated database can process 171 orders per second, which is a 129% improvement. 175 170.6 ORDERS PER SECOND 150 125 100 75 129% more orders per second with PernixData FVP 74.6 50 Virtualized Database (standard) Virtualized Database Accelerated with PernixData FVP PernixData FVP grows revenue by increasing orders processed per second. 3
Order Completion Time is another important metric in measuring performance. The standard database completes the task in 232 ms while the FVP accelerated database finishes orders in 93 ms, reducing the cycle time by 60%. ORDERS COMPLETION TIME (MS) 250 200 150 100 231.5 60% lower latency with PernixData FVP 93.4 50 Virtualized Database (standard) Virtualized Database Accelerated with PernixData FVP PernixData accelerates revenue by decreasing latency in virtualized databases. PernixData FVP accelerates virtualized server to deliver significantly better results in both key metrics compared to a standard virtualized database. It is also important to consider the costs incurred to increase the performance of the virtualized database. In this test, only a mid-range PCIe card and PernixData FVP were added, with no changes to the storage array, at a cost of about $14,000. This combination added a conservative estimate of 125,000 IOPS, making the cost per IOPS about $0.11. In comparison, it would cost approximately $300,000 to get similar performance from a storage array, which is about $2.40 per IOPS. So, not only does PernixData FVP deliver better performance for a virtualized database, it delivers this performance at a cost per IOPS that is impossible to beat. Real World Examples Companies from different industries and of all sizes are implementing PernixData FVP to accelerate the performance of their virtualized databases. Here are some examples: A regional bank headquartered in South Carolina wanted better insight into their financial status. They created a virtualized Microsoft SQL Server data cube that imported multiple data sources for this analysis. The bank allocated 6 hours to complete the job, but they were never able to finish processing all the data in that window. They needed a solution and turned to PernixData FVP to accelerate this process. With FVP in place, the job completes in less than 30 minutes (greater than 12x improvement). Now the bank is aware of what s working, what s not and why, allowing them to improve their daily operations. A religious institution based in the Southeastern United States had an accounting application with a Microsoft SQL Server backend that was running very slowly. Their largest transaction took about 15 minutes to complete. They did a root cause analysis and identified their storage array as the bottleneck. They didn t have the budget to buy a new storage array, so they turned to PernixData FVP. With FVP and server flash in place, the same large transaction now takes under 2 minutes, an 87% reduction in processing time. This has enabled the institution to improve its accounting operations for better employee efficiencies. \ 4
MS SQL Server Response Time (minutes) 16 14 12 10 8 6 4 2 0 Without PernixData 87% Decrease With PernixData PernixData FVP delivers 650% faster database performance A financial services firm in Europe has a virtualized SQL workload that creates reports for clients with strict penalties for not meeting the Service Level Agreements (SLA). The firm was concerned that an explosive growth in data would prevent them from meeting their established SLAs, so they looked to PernixData FVP. They found that when their storage array spikes to 40 to 60 ms of latency, FVP with server Flash keeps database responses to under 3 ms. FVP essentially insulates the SQL Server VM from I/O bottlenecks at the storage array, ensuring that the firm is able to meet their customers reporting expectations. PernixData FVP keeps database queries to under 3 ms even when storage array latency spikes to 40 to 60 ms Summary Databases are among the most I/O intensive workloads in a virtual data center, demanding high data throughput or low I/O latency. Anything less will adversely impact database performance, which in turn can significantly impact business operations. With PernixData FVP and server Flash, companies get the best possible database performance at the lowest cost, regardless of which database (e.g. Oracle, MS SQL Server, MySQL) is used. This makes PernixData FVP a strategic infrastructure component to any database initiative. www.pernixdata.com 1-855-PERNIX-D @PernixData Copyright 2014 PernixData, Inc. All rights reserved. This product is protected by U.S. and international copyright and intellectual property laws. PernixData is a registered trademark and FVP, Flash Hypervisor and Flash Cluster are trademarks of PernixData, Inc. in the United States and/or other jurisdictions. All other brands, products, marks and names mentioned herein may be trademarks or service marks of, and are used to identify, products or services of their respective owners. 5