Research group digest

Cloud providers who operate large and medium-scale data centers and offer their services on a pay-per-usage basis face a challenging problem of running data centers in an energy- efficient manner while ensuring that customers’ expectations in terms of availability are met. Cloud providers benefit from significant economies of scale, since they operate large infrastructures shared [...]

Research group digest

We were recently given a VirtualBox image configured to run the HPCC benchmark suite. The code to determine the OS architecture is pretty naive, and we paid for it. The code simply checked for the presence of the substring ‘amd64′ in the kernel name. The image didn’t have it, so it was migrated to EC2 [...]

Research group digest

This post is about setting up VDE network between qemu machines through a F2F job. Tools I used are dpipe, vde_switch, vde_plug, nc, everything wrapped into python code that is in turn called from F2F job.

  • IPx - machine
  • Sx - vde_switch Network layout:  star-topology - central node is master VM, all slave nodes are connected to its switch using their switches. Master virtual machine configuration: provide IP addresses on VDE. Slave virtual machine configuration: acquire IP from master VM.

F2F job Idea The F2F job defines two possible roles for a host - master or slave. In a master role the host launches master virtual machine which serves as a dhcp server for the VDE network. In a slave role the host launches the slave virtual machine which acquires its IP from master VM and joins VDE network. I chose to use a separate vde_switch in front of each VM regardless of it being a master or slave VM to achieve uniform configuration layout across the network (machine-switch; switch-switch).

VM configuration F2F job handles distributing configuration parameters for virtual machines. An unique id is assigned to each of the slave hosts based on the order of joining the group and in turn each of the slaves configure necessary parameters for launching and connecting the VM. Each slave node needs to know ip and port for VDE switch and ID for calculating unique MAC for the VM. On the host of the master  VM there is set up also a listening socket for each slave VM which is in turn connected to master switch.

Final notes

Finally I want to add some bits I learned about scaling the VDE network using dhcp. In this post the VDE network consists of a single master machine and several slave machines. Because of the boot times for multiple slave machines are different (MAC-s are assigned in random order) then it is tedious to describe all possible situations for interface names in slave's /etc/network/interfaces because by default udev writes rules for MAC-ethX associations. What I did was that I reconfigured udev to forget MAC addresses for certain MAC ranges. It already had rules for not remembering interfaces for virtual machines (wild-card MAC ranges) but they are overridden by the rules for some generic cards (Realtek, etc) which still force them to be written out. After commenting out rules for generic cards, udev didn't remember any MAC-ethX associations and interfaces configuration file for slave machine could be written in a very minimal manner.

Sample F2F job and  helper code can be found here and can be run using the  code from f2f-mobile branch (usual configuration).

Research group digest

Finally I got myself a new laptop that is the fastest thing I've ever used; this significant change in the hardware forced me to run the same test one more time, in case of results being importantly different. So, here are the graphs for the same two things (WIPS and TT.H). The parameters are the same as in the last post, though there is a slight difference: number of emulated browsers is 20 (red) and 40 (green).

Browsing

Ordering

Shopping

As we can see, even though the number of browsers raised two times, the response time and throughput raised about 1.8 times, which is not exact number, but roughly estimated. Next time tests with replicated database will be run.

Research group digest

As organizations increase their offerings of digital services, the demand for computing and storage capability increases. Also the need for faster, more complex data processing is becoming widespread. Additional demand for computing capability can only be met by increasing the processing capacity of servers within data centers. Along with greater computing capability, businesses have increased [...]

Research group digest

Microsoft has recently, in November 2010,  published a white paper with the title “The Economics of the Cloud”. In this white paper the authors mention the following potential cost savings for organisations: Cloud also allows core IT infrastructure to be brought into large data centers that take advantage of significant economies of scale in three [...]

Today I needed to start a couple of long-lived jobs running in VirtualBox instances. The images are 64bit, and the only spare host we have around is Xen with a Debian Squeeze dom0. OK I thought, let’s give VirtualBox on Xen a go. I downloaded the Debian 4.0.4 package from the VirtualBox site and attempted [...]

Reviving Thinkpad X24 with Xubuntu and GLX-Dock

On my experiment turning an old Thinkpad X24 notebook in a nice internet tablet, I have tried out several things. I have still in mind running in the end Android on this device, but I need more time to understand how the Android X86 system works. At the moment it runs in Vesa mode, but this is too sluggish for real use. There seem to be some issues with the radeon module. As an intermediate solution, I decided to use Xubuntu which installed nicely without hassle. I have new fancy very small wlan usb stick (rtl8192cu). This made a little trouble until I tired to compile the drivers available on the cd shipping with it, which worked perfectly with the standard Ubuntu kernel. I had more trouble with suspending the machine. Also here, the radeon module seems to make trouble. After a long search, I found the very simple solution: add radeon.modeset=0 to the kernel command line in grub (edit /etc/default/grub, run update-grub) and install uswsusp. This disbaled the mode switching in the radeon module and now I can suspend -- cool! USWSuspend worked a bit better with my wifi and bluetooth driver. I also added the bluetooth-quirk to as extra parameter /etc/pm/config.d/99myown. The file contains now the following: # my own configs for this here [ulno] SLEEP_MODULE=uswsusp ADD_PARAMETERS="--quirk-bluetooth-service-off"

Now also my little bluetooth dongle works again after a suspend.

I also tried out the kde-netbook-plasma remix to check out th eeyecandy there, but even when using --graphicssystem raster, the system feels to sluggish. Maybe a bit overkill for this old notebook.

However, inspired by kde's eyecandy, I was looking for some more efficient options and stumbled upon glx-dock (former name cairo dock). What a nice clean piece of software! And runs if called with the -c (makiing it use cairo and not opengl) option really quickly. I had to enable compostion manager in xfce, but this is just what the little notebook here can still manage. See yourself:

Looks pretty neat I would say. I think, there is now some life back in this old X24 - maybe I can start thinking now of adding a touchscreen layer to add...