WHITE VMware vSphere™ 4 Fault Tolerance: Architecture and Performance PAPER
VMware white paper Table of Contents 1. VMware Fault Tolerance Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1. Deterministic Record/Replay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2. Fault Tolerance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
VMware white paper VMware® Fault Tolerance (FT) provides continuous availability to virtual machines, eliminating downtime and disruption — even in the event of a complete host failure. This whitepaper gives a brief description of the VMware FT architecture and discusses the performance implication of this feature with data from a wide variety of workloads. 1. VMware Fault Tolerance Architecture The technology behind VMware Fault Tolerance is called VMware® vLockstep.
VMware white paper Figure 2. High Level Architecture of VMware Fault Tolerance Primary Secondary APP OS APP OS FT Logging Traffic VM VM wa re wa re ACKs Record Client Replay Shared Storage The communication channel between the primary and the secondary host is established by the hypervisor using a standard TCP/IP socket connection and the traffic flowing between them is called FT logging traffic.
VMware white paper 2. Performance Aspects and Best Practice Recommendations This section describes the performance aspects of Fault Tolerance with best practices recommendations to maximize performance. For operational best practices please refer to the VMware Fault Tolerance Recommendations and Considerations on VMware vSphere 4 White Paper. 2.1.
VMware white paper To ensure that the secondary virtual machine runs as fast as the primary, it is recommended that: • The hosts in the FT cluster are homogenous, with similar CPU make, model, and frequency. The CPU frequency difference should not exceed 400 MHz. • Both the primary and secondary hosts use the same power management policy. • CPU reservation is set to full for cases where the secondary host could be overloaded.
VMware white paper It is recommended that FT primary virtual machines be distributed across multiple hosts and, as a general rule of thumb, the number of FT virtual machines be limited to four per host. In addition to avoiding the possibility of saturating the network link, it also reduces the number of simultaneous live migrations required to create new secondary virtual machines in the event of a host failure. 2.8.
VMware white paper Figure 4. SPECjbb2005 Performance FT traffic: Relative performance (%) SPECjbb2005 1.4 Mbits/sec 120 100 80 60 FT Disabled 40 FT Enabled 20 0 RHEL 5 64-bit, 4GB 3.2. Kernel Compile This experiment shows the time taken to do kernel compile, which is both CPU and MMU intensive workload due to forking of many parallel processes.
VMware white paper 3.3. Netperf Throughput Netperf is a micro-benchmark that measures the throughput of sending and receiving network packets. In this experiment netperf was configured so packets could be sent continuously without having to wait for acknowledgements. Since all the receive traffic needs to be recorded and then transmitted to the secondary, netperf Rx represents a workload with significant FT logging traffic.
VMware white paper Figure 7. Netperf Latency Comparison FT traffic Netperf - latency sensitive case Rx: 500 Mbits/sec 1000 Tx: 36 Mbits/sec 900 800 Mbits/sec 700 600 500 FT Disabled 400 FT Enabled 300 200 100 0 Receives Transmits 3.5. Filebench Random Disk Read/Write Filebench is a benchmark designed to simulate different I/O workload profiles. In this experiment, filebench was used to generate random I/Os using 200 worker threads.
VMware white paper 3.6. Oracle 11g In this experiment, an Oracle 11g database was driven using the Swingbench Order Entry OLTP (online transaction processing) workload. This workload has a mixture of CPU, memory, disk, and network resource requirements. 80 simultaneous database sessions were used in this experiment. Enabling FT had negligible impact on throughput as well as latency of transactions. Figure 9.
VMware white paper 3.7. Microsoft SQL Server 2005 In this experiment, the DVD Store benchmark was used to drive the Microsoft SQL Server® 2005 database. This benchmark simulates online transaction processing of a DVD store. Sixteen simultaneous user sessions were used to drive the workload. As with the previous benchmark, this workload has a mixture of CPU, memory, disk, and networking resource requirements.
VMware white paper 3.8. Microsoft Exchange Server 2007 In this experiment, the Loadgen workload was used to generate load against Microsoft Exchange Server 2007. A heavy user profile with 1600 users was used. This benchmark measures latency of operations as seen from the client machine. The performance charts below report both average latency and 95th percentile latency for various Exchange operations. The generally accepted threshold for acceptable latency is 500 ms for the Send Mail operation.
VMware white paper 4. VMware Fault Tolerance Performance Summary All Fault Tolerance solutions rely on redundancy. Additional CPU and memory resources are required to mirror the execution of a running virtual machine instance. Also, some amount of CPU is required for recording, transferring, and replaying log events. The amount of CPU required is mostly dependent on incoming I/O.
VMware white paper Appendix A: Benchmark Setup Primary Intel Xeon E5440 2.8GHz – 8 CPUs 8GB of RAM Secondary Cross cable APP OS VM wa re Intel Optin XF SR 10GB NIC Intel NC364T Quadport Gigabit APP OS Intel Optin XF SR 10GB NIC VM wa re ESXi 4.0 ESXi 4.0 Client Intel Xeon E5440 2.8GHz – 8 CPUs 8GB of RAM EMC Clariion CX3-20 Flare OS 03.2.6.020.5.011 AMD Operteron Processor 275 – 4 CPUs 2.21GHz 8GB of RAM Storage Array System: ClariiON CX3-20 FLARE OS: 03.26.020.5.
VMware white paper Appendix B: Workload Details SPECjbb2005 Virtual machine configuration: 1 vCPU, 4GB RAM, Enhanced VMXNET virtual NIC, LSI Logic virtual SCSI adapter OS version: RHEL5.1, x64 Java Version: JRockit R27.4.0, Java 1.6.0_22 Benchmark parameters: No of warehouses: Two J VM Parameters: -XXaggressive -Xgc:parallel -XXcompactratio8 -XXminblocksize32k -XXlargeObjectLimit=4k -Xmx1024m -Xms1024m Note: Scores for the first warehouse run were ignored.
VMware white paper Swingbench configuration: Swingbench version: 2.2, Calling Circle Database No of orders: 23550492 No of Customers: 864967 Runtime: 30 mins ODBC driver: ojdbc6.
VMware white paper Exchange 2007 — Loadgen Virtual machine configuration: 1 vCPU, 4GB RAM, Enhanced VMXNET virtual NIC, LSI Logic virtual SCSI adapter OS version: Windows Server 2003 R2, Datacenter Edition, 64-bit Exchange version: Exchange Server 2007 SP1, 64-bit version (08.01.0240.006) Exchange configuration: AD, Mail Hub, IIS and all other Exchange components installed on the same virtual machine Exchange Database: Two 150GB databases, each hosting 800 users Loadgen version: 08.02.
VMware, Inc. 3401 Hillview Ave Palo Alto CA 94304 USA Tel 877-486-9273 Fax 650-427-5001 www.vmware.com Copyright © 2009 VMware, Inc. All rights reserved. This product is protected by U.S. and international copyright and intellectual property laws. VMware products are covered by one or more patents listed at http://www.vmware.com/go/patents. VMware is a registered trademark or trademark of VMware, Inc. in the United States and/or other jurisdictions.