PARTNER PERFORMANCE BENCHMARK PAPER / SANDISK Marvell DragonFly With SanDisk X110 SSD 256GB TPC-C Benchmark Test Results TPC-C OLTP Database Benchmark: 20x Higher-performance using Marvell DragonFly NVCACHE with SanDisk X110 SSD 256GB JUNE 2013
CONTENTS Marvell DragonFly with SanDisk X110 SSD 256GB TPC-C Benchmark Test Results 1 DOCUMENT OVERVIEW...3 2 TEST ENVIRONMENT...3 3 DATABASE CONFIGURATION AND TEST METHODOLOGY...4 4 TEST RESULTS...5 5 SUMMARY... 7 2
1. DOCUMENT OVERVIEW The goal of this document is to describe how SanDisk X110 SATA Solid-State Drives (SSD) 256GB, along with the Marvell DragonFly Caching solution, can significantly improve database performance. The approach selected to generate the results is to utilize an open-source TPC-C Benchmark Test Suite (Hammerora) using a MySQL 5.5 back-end database. There are two TPC-C critical metrics gathered on this performance evaluation and they are (Transactions Per Minute) and (Number of Orders Per Minute). This test compares the cumulative and numbers when the database resides entirely on a Dell EqualLogic 10GbE iscsi back-end data store configuration versus when the database resides on Marvell DragonFly cache consisting of four SanDisk X110 SSD 256GB, utilizing the same Dell EqualLogic 10GbE iscsi as back-end data store. 2. TEST ENVIRONMENT Below are the hardware and software sets utilized to perform the performance evaluation: BACK-END DATA STORE CONFIGURATION Server Dell EqualLogic PS6110 Storage 14 1TB Dell SATA Drives Networking Dual 10 GbE (in Failover mode) HOST SERVER CONFIGURATION Server Brand and Model Dell Compellent SC8000-R720 Processors 12 Core Intel Xeon E5-2640 2.5GHz Memory 40 GB Operating System RHEL 6.2 Marvell DragonFly Card 1 Marvell DragonFly Software R1.1.1 Marvell DragonFly attached SSDs 4 x 6Gb/s SanDisk X110 SATA SSD 256GB 10 GbE Network Adapter Chelsio N320E NETWORK INFRASTRUCTURE 10 GbE Switch Cisco Nexus 5100 3
Testing MySQL on Marvell DragonFly Setup Example Hammerora TPC-C 100 users, 100 Warehouses x 8 VMs [16 Core 2 Socket Xeon/ 40GB RAM] [16 Core 2 Socket Xeon/ 40GB RAM] iscsi over 10GbE Marvell DragonFly No Cache Setup 10GbE Switch PCI-x8 Slot iscsi Array Up to 4 SAS or SATA SSDs Figure 1. MySQL on Marvell DragonFly Setup. To compare the benchmark numbers, two types of configurations were created: 1. No Cache Server Configuration Hammerora running directly on Dell EqualLogic 10 GbE back-end data store (/dev/sdh mounted as /mnt/iscsi_storage). 2. With Marvell DragonFly Cache Server Configuration Hammerora running on Marvell DragonFly Cache (/dev/mvsablk_0 mounted as /mnt/cached_iscsi_storage) consisting of four SanDisk X110 SSD 256GB, using the same Dell EqualLogic 10 GbE iscsi for back-end data store. 3. DATABASE CONFIGURATION AND TEST METHODOLOGY This diagram describes how Hammerora (an open-source database load test tool) was used to generate TPC-C type of workload. The database workload configured consists of eight Hammerora Clients and eight MySQL databases installed on 8 RHEL 6.2 Virtual Machines. Hammerora Test Methodology Hammerora Application TPC-C test: 100 users 100 Warehouses 30 minutes on each VM VMs: Each running Hammerora MySQL Ext3 on RHEL 6 MySQL Database /dev /vda KVM on RHEL 6 QCOW2 Master QCOW2 Clone QCOW2 Clone QCOW2 Clone QCOW2 Clone Block Dev With Ext3 FS Figure 2. Hammerora Test Methodology. 4
4. TEST RESULTS The bar graphs below show the cumulative and numbers obtained after running the Hammerora Benchmark Tests on TPC-C benchmark configurations without and with Marvell DragonFly. No Caching With Marvell DragonFly Cache 4441844 220626 No Caching With Marvell DragonFly Cache 67973 3370 Figure 3. Cumulative and with No Cache and With Marvell DragonFly Cache The table below shows the and observed per TPC-C Database instance in each VM. NO CACHING 4 SANDISK X110 SSD 256GB FOR CACHING WITH MARVELL DRAGONFLY VM1 26,484 406 512,046 7,835 VM2 26,097 399 541,571 8,286 VM3 27,311 418 590,615 9,036 VM4 31,631 481 577,606 8,838 VM5 28,719 439 556,771 8,511 VM6 26,557 405 574,077 8,806 VM7 26,373 403 549,355 8,395 VM8 27,454 419 539,803 8,266 Totals 220,626 3,370 4,441,844 67,973 5
5. SUMMARY In summary, this benchmark illustrates how Marvell DragonFly cache is a cost-effective way to significantly improve database performance. As shown in the table below, when all the s and s are added, an approximately 20x improvement is exhibited when Marvell DragonFly caching is enabled. METRIC IMPROVEMENT 20x 20x For more information about Marvell DragonFly go to: http://www.marvell.com/storage/dragonfly/. ABOUT MARVELL: Marvell (NASDAQ: MRVL) is a global leader in providing complete silicon solutions enabling the digital connected lifestyle. From mobile communications to storage, cloud infrastructure, digital entertainment and in-home content delivery, Marvell s diverse product portfolio aligns complete platform designs with industry-leading performance, security, reliability and efficiency. At the core of the world s most powerful consumer, network and enterprise systems, Marvell empowers partners and their customers to always stand at the forefront of innovation, performance and mass appeal. By providing people around the world with mobility and ease of access to services adding value to their social, private and work lives, Marvell is committed to enhancing the human experience. As used in this release, the term Marvell refers to Marvell Technology Group Ltd. and its subsidiaries. For more information, please visit www.marvell.com. Marvell Semiconductor, Inc. 5488 Marvell Lane Santa Clara, CA 95054 Phone 408.222.2500 www.marvell.com Copyright 2013. Marvell International Ltd. All rights reserved. Marvell and the Marvell logo are registered trademarks of Marvell. DragonFly is a trademark of Marvell or its affiliate. All other trademarks are the property of their respective owners. Marvell_DragonFly_SanDisk-02_benchmark paper 6/13 6