Виртуализация
Методика и результаты тестирования масштабируемости гипервизора Red Hat при виртуализации баз данных

Методика и результаты тестирования масштабируемости гипервизора Red Hat при виртуализации баз данных

Отчет о тестировании Virtualized database hypervisor scaling report: Red Hat Kernel. Virtual Machine running on the Dell PowerEdge R710 solution. TEST REPORT. AUGUST 2009 включает методику и результаты запуска 8 виртуальных машин с MySQL СУБД на гипервизоре Red Hat KVM в бета-версии RHEL 5.4.

Результаты, естественно, устарели, но отчет ценен подробным описанием методики тестирования.

image

Тест:

We ran the KVM hypervisor solution on a Dell™ PowerEdge™ R710 server and two shelves of Dell EqualLogic™ PS5000XV storage. The Dell PowerEdge R710 server had two 2.66GHz Intel® Xeon® Processor X5550s and 72 GB of RAM. We connected the Dell PowerEdge R710 server to the Dell EqualLogic storage via a 10Gb iSCSI connection. We installed Red Hat Enterprise Linux 5.4 Beta (kernel-2.6.18-159.el5) as the host operating system, and installed Red Hat Enterprise

Linux 5.4 Beta (kernel-2.6.18-155.el5) as the guest operating system. For this scaling analysis, we selected the DVD Store Version 2 (DS2) test tool. DS2 is an open-source simulation of an online e-commerce DVD store, where customers log in, browse, and order products. Each virtual machine (VM, also known as “guest”) ran a single MySQL workload. One copy of the DS2 database was on each VM, with a 4GB database per VM. By combining multiple virtual machines, all running a CPU-, memory-, and disk-intensive workload, the test showed the scaling ability of the Red Hat Enterprise Linux with KVM platform, especially as it relates to an active OLTP environment. The main DS2 metric is orders per minute (OPM). We report OPM results from a 300-second period of steady activity and heavy load during the DS2 test run. The Workload section further explains the MySQL database workload.

Результаты

    • While running one VM, the KVM solution delivered 14,204 OPM;
    • while running two VMs, the KVM solution delivered 28,219 OPM (1.99 times more than 1 VM);
    • while running four VMs, the KVM solution delivered 49,826 OPM (3.51 times more than 1 VM);
    • while running eight VMs, the KVM solution delivered 77,197 OPM (5.43 times more than 1 VM)

1 thought on “Методика и результаты тестирования масштабируемости гипервизора Red Hat при виртуализации баз данных

    • Author gravatar

      Аналогичный тест для виртуализации Red Hat MRG
      http://www.principledtechnologies.com/clients/reports/Red%20Hat/RHEL_KVM_AMQP.pdf

      Virtualized messaging hypervisor analysis on Red Hat Kernel Virtual Machine running on the Dell PowerEdge R710 solution TEST REPORT SEPTEMBER 2009

      KEY FINDINGS

      While running one- and two-guest virtualized latencytest messaging workloads, the KVM solution delivered sub- 200-microsecond latencies at one thousand messages per second between the client system and the Dell PowerEdge R710 at all but the largest message size. (See Figure 1.)

      While running two simultaneous virtualized perftest messaging workloads, the KVM solution delivered over one million messages per second at 16-byte and 32- byte message sizes, and delivered over 800 MB/s at larger message sizes.

      Executive summary

      Red Hat, Inc. (Red Hat) commissioned Principled Technologies (PT) to analyze the virtualized messaging performance of the Red Hat® Kernel Virtual Machine® (KVM) hypervisor with VMs running Advanced Message Queuing Protocol (AMQP). We ran Red Hat Enterprise Linux (RHEL) with KVM on a Dell™ PowerEdge™ R710 with two 2.93GHz Intel® Xeon® Processor X5570s and 24 GB of RAM.

      We installed Red Hat Enterprise Linux 5.4 Beta (kernel-2.6.18- 159.el5) as the host operating system and Red Hat Enterprise Linux 5.4 Beta (kernel-2.6.18-159.el5) as the guest operating system. For this analysis, we used the perftest and latencytest tools, which Red Hat includes with MRG Messaging, to compare both throughput and latency on the KVM hypervisor platform. We ran each test in both one- and two-guest configurations. Latencytest scores measure response times, making lower numbers better, while perftest scores measure number of messages transmitted successfully per second, making higher numbers better. The Workload section explains the perftest and latencytest workloads in greater detail.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.