White Paper EMC XtremSF: Delivering Next Generation Performance for Oracle Database Abstract This white paper addresses the challenges currently facing business executives to store and process the growing volumes of data driven by such things as new applications, virtualization, the cloud and explosive growth in web consumerization. In this paper, EMC demonstrates how advanced storage technology can be applied to Oracle database enterprise environments to greatly improve performance and reliability while lowering costs. March 2013
Copyright 2013 EMC Corporation. All Rights Reserved. EMC believes the information in this publication is accurate as of its publication date. The information is subject to change without notice. The information in this publication is provided as is. EMC Corporation makes no representations or warranties of any kind with respect to the information in this publication, and specifically disclaims implied warranties of merchantability or fitness for a particular purpose. Use, copying, and distribution of any EMC software described in this publication requires an applicable software license. For the most up-to-date listing of EMC product names, see EMC Corporation Trademarks on EMC.com. H11578 2
Table of Contents EXECUTIVE SUMMARY... 4 EMC XTREMSF FLASH STORAGE TECHNOLOGY... 5 KEY FINDINGS... 6 ONLINE TRANSACTION PROCESSING TEST... 7 Traditional DAS Performance Compared to XtremSF... 7 Figure 1: Oracle OLTP Comparative Performance Benchmark... 8 Oracle Data and Log Location with XtremSF... 8 Figure 2: Performance Effect of Separating Oracle Data and Log with XtremSF... 8 Oracle OLTP on Hybrid solution with XtremSF... 9 Figure 3: Hybrid deployment model... 9 Figure 4: Oracle OLTP with Hybrid Storage for Data & Log... 10 ORACLE ASM MIRRORING FOR HIGH AVAILABILITY... 11 Figure 5: OLTP Performance for Oracle ASM Mirroring with XtremSF and VNX Array... 11 ORACLE DATA GUARD TEST... 12 Figure 6: Oracle Data Guard with XtremSF Configuration... 12 Figure 7: Oracle Data Guard with Hybrid Storage Configuration... 13 CONCLUSION... 13 3
Executive Summary The enormous volume of data stored by businesses is growing exponentially, increasing system workloads and creating new challenges for IT organizations. At the same time, organizations are delivering expanded software capabilities to users that increase job complexity, all the while demanding faster results. The primary challenge for enterprise environments looking to meet this new digital data paradigm is to overcome the limitations of traditional storage related to rotational latency. This is the time it takes for the rotation of the disk to bring the required disk sector containing the requested data under the read-write head. The demand for greater I/O performance driven by such things as new applications, virtualization, the cloud and an explosive growth in web consumerization has created a challenge for many companies. While processor speeds and network bandwidth advancements have kept pace with enterprise requirements, data storage I/O (input/output) has remained a key performance constraint. As increased data load requirements continue to outpace traditional storage I/O technology, businesses are challenged with how to accelerate performance to meet new increased transactional processing and data analysis loads. The emergence of flash storage technologies offers orders-of-magnitude faster I/O performance, but there are differing approaches that need to be considered. The purpose of this White Paper is to provide the results of tests conducted by EMC that compared the performance of local server storage, XtremSF - a new advanced Flash storage technology - and a mix of traditional storage array and XtremSF within an Oracle database enterprise environment. These results showed an unprecedented price/performance gain when running Oracle by delivering maximum throughput performance while reducing infrastructure cost and lowering power consumption and cooling requirements. This White Paper contains the following: An introduction to EMC XtremSF flash storage technology; Key findings from benchmark tests conducted and configurations to achieve optimal performance; and Conclusions This white paper is intended for Senior Management, IT Management, and IT Planners and Database Administrators. 4
EMC XtremSF Flash Storage Technology EMC XtremSF is an advanced flash storage technology deployed in a server PCIe card slot designed to deliver unprecedented application performance by reducing latency and accelerating I/O throughput. XtremSF has been designed to treat flash media like an extension of memory while maintaining a traditional block storage interface for applications. It allows applications to access data in the most efficient manner possible. Residing on the server PCIe interconnect bus, XtremSF overcomes the limitations of mechanical disk performance and the bandwidth constraints of network storage access - reducing response time from milliseconds to microseconds. By storing data locally in the server, XtremSF alleviates I/O bottlenecks to deliver maximum read and write performance. When measured at near-full capacity, XtremSF delivers two times greater performance than its closest competitor. Additionally, XtremSF offers the industry s highest storage capacity in the smallest footprint. At half-height, halflength, XtremSF is a single low-profile card that fits in any rackmount server within the power envelope of a single PCIe slot. Its compact, universal form factor enables organizations to maximize the use of valuable datacenter space - reducing data center sprawl. XtremSF is available in 550 GB, 700 GB, 1.4 TB, and 2.2 TB MLC Flash based storage capacities, in addition to 350GB and 700GB SLC Flash based PCIe cards. XtremSF delivers sustained, predictable, high performance across a variety of mission-critical application and database workloads. XtremSF is particularly ideal for high-transactional and high-performance workloads often associated with the following operational environments: High-performance trading applications Web 2.0 applications High-performance computing (HPC) This advanced storage technology can also be used to accelerate analytics, reporting, data modeling, indexes, database dumps, batch processing, background tasks, and other ephemeral workloads. With XtremSF, both read and write database operations are performed directly to the PCIe flash card in the server. XtremSF delivers performance without compromising over the entire lifetime of the device, even when the device is full. To maximize the lifetime of the flash media, XtremSF performs dynamic global and local 5
wear leveling; whenever needed, XtremSF relocates data to areas of flash that are less used. Furthermore, sophisticated scheduling algorithms allow flash management tasks to occur without impacting application performance. XtremSF is specifically designed to minimize CPU overhead in the server by offloading these flash management operations from the host CPU to the PCIe card. XtremSF offers simplified management. Unlike competing solutions, 100% of the XtremSF capacity is available as a single host volume on the server. XtremSF also delivers continuity of operations in the event that there is a failure within any flash media component. Inherent in XtremSF, flash components can be isolated as separate slices and spread across RAID groups, allowing for multiple failures to occur without disrupting an application s access to the data. Organizations that want accelerated performance with protection can combine XtremSF with EMC VFCache - intelligent server flash caching software. While accelerating reads, VFCache simultaneously delivers enterprise-class protection of mission-critical application data. The software caches an application s most frequently accessed data on a PCIe flash card in the server to deliver maximum performance. Furthermore, it leverages a write-through algorithm to ensure that newly written data matches the networked storage array for persistent high availability, integrity, reliability, and disaster recovery. Key Findings EMC s overall findings include performance results that also demonstrate best practices when deploying multiple storage technologies within your storage architecture to deliver optimal performance at the lowest cost. These findings were collected running Oracle 11g R2 under the Redhat 5.7 operating environment. Performance testing focused on the following three application areas: 1. High Volume Transaction Processing; 2. High Volume Transaction Processing with Oracle ASM Mirroring for HA 3. OLTP with Oracle Data Guard. Furthermore, by deploying XtremSF particularly in a high volume online transaction environment results suggest that organizations are able to lower overall CAPex (capital expenditure) and OPex (operational expenditure) in the following ways: 6
1. XtremSF greatly minimized the number of mechanical drives required to achieve similar optimal performance. This reduced infrastructure should also translate into reduced cooling and power requirements and lower TCO and maintenance costs; and 2. XtremSF frees server resources, thereby, allowing a single server to handle a greater workload serving multiple databases or applications. This reduced infrastructure should also translate into reduced cooling and power requirements and lower TCO and maintenance fees. Online Transaction Processing Test There are three parts to this test running Oracle in a high transaction processing (TPC-C like) model environment. For the first and second part of this benchmark, EMC established the following configuration: Server: Cisco C-460 M2 with 512GB memory Operating System: Redhat 5.7 Storage: 10 x 600GB SAS (serial attached SCSI) drives 1 x XtremSF 2.2TB for DATA 1 x XtremSF 550GB for LOG Database: Oracle 11g R2 populated with 1.2TB of TPC-C like OLTP data Traditional DAS Performance Compared to XtremSF In this first test, EMC found that with direct attached storage the benchmark was able to achieve 4,300 tpm (transactions per minute). After replacing the direct attached storage with single XtremSF card hosting both DATA and LOG, transaction processing performance increased to nearly 115,000 tpm. This XtremSF result was an incredible 2,600% performance advantage when compared to a direct attached storage alternative (Figure 1). 7
Figure 1: Oracle OLTP Comparative Performance Benchmark Transactions per minute 140000 120000 100000 80000 60000 40000 20000 0 4,300 Local Drives 115,018 VFStore Oracle Data and Log Location with XtremSF In this test, EMC tested the impact on OLTP performance when combining the DATA and LOG on a single XtremSF PCIe card compared to dedicating each one to its own XtremSF PCIe card (i.e., two XtremSF cards). As indicated in Figure 2, the results of this test showed that separating the Redo Log and Data on to separate XtremSF cards reduced the log file sync & log file parallel write latencies ultimately increasing performance by 42%. Figure 2: Performance Effect of Separating Oracle Data and Log with XtremSF Transactions per minute 180000 160000 140000 120000 100000 80000 60000 40000 20000 0 162,814 DATA & LOG on Separate VFStore Cards 115,018 DATA & LOG on Same VFStore Card 8
Oracle OLTP on Hybrid solution with XtremSF The traditional storage arrays like EMC VMAX and VNX provide capacity at scale and also several other data services. Combining the performance advantages of XtremSF with capacity benefits of traditional storage arrays provides very low TCO for the Customer. For our third test, EMC established the following hybrid storage configuration: Server: Cisco C-460 M2 with 512GB memory Operating System: Redhat 5.7 Network: 2 x Brocade 5100 Fibre Channel Storage: VNX 7500-10 x 600GB SAS drives 1 x XtremSF 2.2TB for Data Database: Oracle 11g R2 with 1.2TB of TPC-C like OLTP data Figure 3: Hybrid deployment model for Oracle OLTP The purpose of this test was to determine the effect on high volume transaction performance by splitting the Oracle Data and Log files between XtremSF and an EMC VNX storage array. The Oracle Data was placed on a single XtremSF PCIe card and the Log files were placed on a VNX storage array. The results of this test showed that by combining 9
XtremSF, with traditional storage arrays users can leverage both performance advantages of XtremSF and capacity provided by traditional storage arrays. This hybrid storage solution delivered over 150,000 tpm. This test result delivered a 31% improvement compared to placing both data and log files on a single XtremSF card although about 7.5% lower compared to separating these files on two XtremSF cards (Figure 4). Beyond performance, another key point to consider with this hybrid storage solution is that traditional storage arrays are able to provide higher capacity for storing several days worth of archive logs. If the user is unable to host the entire database data on XtremSF cards, user can identify their IO objects with high IOPs requirement like database indices and move them to XtremSF. Figure 4: Oracle OLTP with Hybrid Storage model 160000 150,584 Transactions per minute 140000 120000 100000 80000 60000 40000 20000 115,018 0 DATA & LOG on one VFStore Card DATA on VFStore & LOG on VNX (10 SAS) 10
Oracle ASM Mirroring for High Availability For this benchmark, the same configuration was used as in the previous test. The purpose of this test was to simulate a high volume transaction processing environment with Oracle ASM mirroring for high availability within hybrid storage architecture (i.e., XtremSF and VNX array). In this test, Oracle ASM Preferred Read was turned off to establish a tpm baseline. Next, Preferred Read was turned on and set to the XtremSF flash storage device. Results showed that transaction performance more than doubled (Figure 5). This is due to the fact that without preferred Read set to XtremSF, I/O read performance was limited to the speed of the slower I/O from storage array. In this example, setting Preferred Read ensured that Oracle ASM reads would take advantage of the higher I/O performance of XtremSF. The performance gains realized by applications in this hybrid model with ASM preferred read relies on how well the back-end storage array can scale with increased number of writes resulting from improved transactions after deploying XtremSF. To leverage the full potential of XtremSF performance, user can use technologies like FAST Cache or FAST VP to improve the write performance of the back-end storage array. Figure 5: OLTP Performance for Oracle ASM Mirroring with XtremSF and VNX Array 2.5 2.17 Relative Transactions per minute 2 1.5 1 0.5 1.00 0 Without ASM Preferred Read With ASM Preferred Read set to VFStore 11
Oracle Data Guard Test XtremSF provides superior application performance compared to its competition. At the same time XtremSF like any other PCI-e DAS storage lacks high availability features that are provided by traditional storage arrays like EMC VMAX & VNX. One way to benefit from the outstanding performance delivered by XtremSF and overcome the limitation of high availability is to take advantage of Oracle Data Guard, an integrated high availability and disaster recovery feature available in Oracle. Of course, IT enterprises understand that implementing a high availability and disaster recovery solution comes at a cost to performance due to a number of factors. These factors include the type of replication, the volume of log shipping, the distance between the primary and secondary sites and the quality and speed of the interconnect or log replication network between sites. One approach is to implement XtremSF as the storage solution at both the primary/production site and also at the disaster secondary/recovery site (Figure 6). Figure 6: Oracle Data Guard with XtremSF Configuration A second approach is to implement XtremSF as the storage solution for the primary site to take advantage of the higher performance capabilities of this solution and to use a VMAX or a VNX storage array for the secondary site (Figure 7). 12
Figure 7: Oracle Data Guard with Hybrid Storage Configuration There are several advantages to this hybrid approach. 1. A powerful EMC storage array can act as a secondary destination for multiple sites or applications. (Assumption here is that, not every application will have outage at the same time) 2. The storage array cloning features can be used on the secondary site for creating test/development copies of the database 3. The storage array on the secondary site can also be used for performing reporting tasks 4. The secondary site can be used to offload backup task Conclusion As IT planners prepare their data center to meet the growing need to store increasing volumes of data and to deliver greater I/O performance in a cost-effective manner, they should look to a storage architecture that combines EMC VNX Arrays with XtremSF flash technology to deliver immediate value. Test results with Oracle 11g demonstrate that an architecture that uses this storage solution can dramatically improve I/O performance and provide opportunities to reduce overall CAPex and OPex through increased performance for high volume transaction processing, a highly efficient storage architecture, lower cooling and power requirements and reduced maintenance costs. 13
With XtremSF, IT organizations can immediately realize unprecedented gains in database and application performance for demanding enterprise environments that include: High-performance trading applications Web 2.0 applications High-performance computing (HPC) The benefits from this advanced storage solution approach are revolutionizing the way leading IT organizations should be thinking about their future server and storage architectures in order to meet the new data processing and performance requirements of today s demanding enterprise environments. 14