Stress testing

This test is made to stress core, middleware plugins and underlying virtual infrastructure. To make this we send lots of virtual machines so see how ONE behaves with big loads.

For this tests we have configured 4 machines with xen and no nfs shared storage. Virtual machines sent are little ttylinux images of 5Mb that have 128 Mb of ram allocated. No network interfaces are created for this virtual machines. 30 of this machines are able to run at a time in each of the machines.

400 VM's are sent to one and the scheduler looks for more VM's to run each 30 seconds. There is also a script that looks for already running VM's and sends them shutdown so new VM's can be sent to the nodes.

Transfer manager is not only copying virtual image to the destination host but also creating 1Gb swap images that are attached to each VM.

The VM description used is this one:

NAME = ttylinux

CPU    = 0.25
MEMORY = 128

OS = [
  kernel   = "/vmlinuz",
  initrd   = "/initrd.img",
  root     = "sda1" ]

DISK = [
  source   = "ursa:/images/playground/jfontan/ttylinux/ttylinux.img",
  target   = "sda1",
  readonly = "no" ]

DISK = [
    type="swap",
    size=1024,
    target="sdb"
]

Disk image is located in server ursa and Transfer Manager copies using scp that file to the corresponding host. Swap image is automatically generated by the Transfer Mamager.

This test is similar to the previous one but now it starts 120 machines, waits for all to be running and shuts down all the machines so we can measure times when lots of operations are being done.

Computers that run VM's have these characteristics:

• 2 Quad core Xeon processors, 2Ghz
• 8Gb of ram
• 1Gb copper ethernet interconnection
• Ubuntu Server 8.04
• Xen hypervisor version 3.2