New performance benchmark record for VMware vSphere

Efficiency of VMware server consolidation showed with performance benchmark record.

The benchmarking was done by ParAccel, Inc., provider of ParAccel Analytic Database (PADB), vSphere 4 delivered the industry’s first fully-audited TPC-H benchmark result in an x86 based virtual environment.

A cluster of 80 virtual machines running PADB, spread across 40 physical hosts (40 HP DL380 blades), broke the existing world records for performance and price. New  established a record is 1,316,882 Composite Queries per hour. $ 0.70 Query-per-Hour Performance. The TPC-H Price/Performance metric is expressed as $/QphH@Size.

Update:  Scott Drumond – the man who organized and coordinated the testing – said:

The testing was done on-site at ParAccel with the help of four database performance engineers from VMware. Together with Chethan Kumar, a member of a performance team in VMware and who served as the corporate-based results validation for the project by analyzing esxtop logs and providing recommendations.  I have similarly been assisting with analysis and guidance and have been publicizing the event.

VMware vSphere 4 enables efficient utilization of hardware resources through consolidation or physical servers and achieves this without compromising on performance.

Note that the Hypervizor beeing used was not the “full blown” ESX 4 but the tiny ESXi 4 which shares the same code with ESX 4, but lucks the Red Hat service console.

Here are some of the strong points from the event:

  • Ran 1,316,882 Composite Queries per Hour (QphH) @ 1,000GB, 13% higher than the next best performance.
  • Used 40 HP servers, 37% fewer than that used for the second best performance result.
  • Achieved a price/performance of US $0.70 per QphH, 7.7 times better than that of the previous performance record holder.
  • Took 16 minutes and 23 seconds to load 1 TB database at a rate of 3.7 TBs per hour, a load time 8.7 times  faster than that of the previous performance record holder.

The material and Hypervizor:

Servers:
40 x HP DL 380 G6 Server, each with

  • 2 Intel Xeon x5560, 2.8GHz processors ( 4 cores per socket)
  • 72 GB Memory
  • 8 x 300 GB (10K RPM) internal SAS Disks
  • 2 x HP PCIe Dual Port Gigabit network adapters

Hypervisor:
VMware, ESXi 4.0 update 1

Virtual Machines:
Leader Node:

1 virtual machine with

  • 4 vCPU
  • 32 GB memory
  • 4 x 279 GB virtual disk
  • 5 x virtual NICs
  • ParAccel Standard Linux, 64-bit
  • ParAccel Analytics Database, ver 2.5

Compute Node:
79 virtual machines with

  • 4 vCPU
  • 32 GB memory
  • 4 x 279 GB virtual disk
  • 4 x virtual NICs
  • ParAccel Standard Linux, 64-bit
  • ParAccel Analytics Database, ver 2.5

The results in details on TPC.org

Source: InformationWeek and VMware VROOM! blog

Share

Technorati Tags: , ,

3 Comments

  1. i don’t think Scott did the testing by the way :-)

Trackbacks and Pingbacks

Leave a Reply

Other posts from Server Virtualization