An Ubuntu Enterprise Cloud may consist of as few as two physical systems, up to potentially hundreds or perhaps thousands.

This page provides some basic suggested topology recommendations, based on the number and type of physical systems you have available for use in your cloud.

Note: You may add as many nodes (NCs) to any of these topologies as you wish.

Note: Please refer to UEC/Glossary for explanation of terms used in this page.

This is the most basic possible configuration, with the CLC/Walrus/CC/SC and VM hosting (NC) all on the same machine. This is no different to any other mode, except that network traffic for the node controllers does not have to traverse the local network. There have been some bugs found (and already fixed) when installing in this environment ( 503180 , 519648 , 519653 )

  1. machine A: CLC/Walrus/CC/SC/NC

Sample configuration (existing LAN)

LAN networking configuration:

  • Network
  • DHCP and DNS servers
  • UEC server
  • /etc/network/interfaces on the UEC server

    iface eth0 inet manual
    auto br0
    iface br0 inet static
        # dns-* options are implemented by the resolvconf package, if installed
        # these are google's dns servers
        # this setting could be "local" if you all your local hosts
        #  were named something like a.localdomain b.localdomain c.localdomain etc
        dns-search localdomain
        bridge_ports eth0
        bridge_fd 9
        bridge_hello 2
        bridge_maxage 12
        bridge_stp off
  • /etc/eucalyptus/eucalyptus.conf - key elements to check:

    VNET_PUBINTERFACE="br0" # must match the br definition above
    VNET_PRIVINTERFACE="br0" # must match the br definition above
    VNET_SUBNET="" # ips from this range are bound to $VNET_PRIVINTERFACE:priv
    VNET_DNS="" # what DNS server nodes are given
    # v - should be ips on the LAN, not in use anywhere, and excluded from the DHCP server

If installing by following UEC/PackageInstall you will normally end up doing the br0 configuration *after* installing the cluster controller service and thus end up with eth0 rather than br0 in the eucalyptus.conf file. So be sure to check. If it is wrong, you need to:

sudo stop eucalyptus CLEAN=1
sudo /etc/init.d/eucalyptus-nc stop
# fix the config file here
sudo start eucalyptus CLEAN=1
sudo /etc/init.d/eucalyptus-nc start

The CLEAN parameter makes it rebuild part of a hidden database file with the resulting configuration. Check ip addr to make sure that nothing is bound to eth0 - if it is the easiest way to be sure you have a clean config is to reboot and check again. If the PRIV interface is wrong, you may see nodes get DHCP replies from the LAN DHCP server rather than the cc private DHCP server, and things will be messy.

It should be possible to run on a completely self contained system too, but no one has documented that (except perhaps using SYSTEM mode, but that disables some key features like the 'user data' scripts). Running on a laptop using network manager roaming should also be possible, as long as the machine is completely self contained when disconnected - it needs its own DNS server, nat rules for the 'public ip range' etc.

At least 2 Physical Systems

This configuration puts all of the user facing components (CLC/Walrus) and back-end control components (CC/SC) on a single system, and uses the second for VM hosting.

  1. machine A: CLC/Walrus/CC/SC

  2. machine B: NC

  3. (more NCs...)

At least 3 Physical Systems

This configuration puts user facing components (CLC/Walrus) on one machine, back-end control components on another (CC/SC), and uses the third for VM hosting (NC).

  1. machine A: CLC/Walrus

  2. machine B: CC/SC

  3. machine C: NC

  4. (more NCs...)

At least 4 Physical Systems

This configuration puts user facing components on their own machines, one for user/VM control (CLC) and one for storage (Walrus). The third machine is for back-end control (CC/SC) and the fourth for VM hosting (NC)

  1. machine A: CLC

  2. machine B: Walrus

  3. machine C: CC/SC

  4. machine D: NC

  5. (more NCs...)

At least 5 Physical Systems

This configuration puts user facing components (CLC/Walrus) on one machine, back-end control components on another (CC/SC), and uses the third for VM hosting (NC), fourth for a second CC/SC, and fifth for an additional NC using resources from CC2/SC2.

  1. machine A: CLC/Walrus

  2. machine B: CC1/SC1

  3. machine C: NC1

  4. machine D: CC2/SC2

  5. machine E: NC2

  6. (more CCs/SCs/NCs...)

UEC/Topologies (last edited 2010-07-24 03:27:07 by ideaman007)