Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Best strategy to make Cloud using 2 Physical server
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Best strategy to make Cloud using 2 Physical server

indiavpsindiavps Member
edited November 2015 in General

Can some one advice the best strategy to make Cloud using 2 Physical servers (with minimal additional resource) and with a possibility to expand.

Comments

  • Use cloud software like Kubernetes, Openstack, Cloudstack, Cloudfoundry, OpenNebula. All of them can be easily expanded. If you need to add a SAN storage server in the future, there isn't any issues if you go with Openstack.

    Thanked by 2vpsGOD indiavps
  • It very much depends on what you are trying to achieve (i.e. what you mean by "cloud" which is a term used to cover many things)...

  • @GIANT_CRAB
    Thank you, I was think Openstack, Cloudstack or May be Ubuntu EC (Eucalyptus powered). Do you feel any critical advantage among any ?

    @MeAtExampleDotCom
    Thanks, you can consider it like a private cloud and run hypervisor.

  • for "real" cloud you need 4 servers + 2 switches:

    HN1

    HN2

    Storage1 (ST1)

    Storage2 (ST2)

    Switch1 (SW1)

    Switch2 (SW2)

    Connect ST1+2 via bonded (or failover, but no single link) 10GE+/Infiniband/FC (or GE if you really have no other choice) to each other or 2 redundant switches (splitting 2 switches in 2 VLANs each for storage/HN redundant is fine, no need for 4 total).

    Connect HN1+2 to same switch, again redundant - Silver edition: use bonded GE/10GE for network and bonded 10G+ for storage, prevents congestion issues due to load of each other.

    Install ceph on storage (or glusterfs on top of like ZFS) and add them to be sync (master+master essentially), use iSCSI to mount the volumes on HN (via hypervisor best case, not via Iscsiadm locally), install KVM/Xen (OVZ needs more work as it requires a filesystem on the share), point HV to iSCSI share and configure either watchdog (inside KVM or on host) or something similar (Xen has their own implementation), sync config and check other node (laziest method by ping, better by local daemon called by the other node) - if down kill the storage (switch)ports to the other node (to prevent split brain once it comes back) and boot up the VMs on HN2.

    Premium edition: Configure both servers to use RAM mirroring internal (halfs your total RAM) to prevent the rare ECC failures ("RAID5" after all). If your servers can do it (very unlikely, that is an E7 feature, iirc not even E5 has it) you can also do CPU mirroring.

    Ultra Premium edition: Move RAM/CPU cache from HN1 to HN2 live by keeping it in sync, requires 100GE between them to work and adds 500%+ ram latency (kernel and network overhead) with 25%+ (depending on RAM tech and network bonding, 100GE does ~10GB/s p. link while DDR3 does ~6.5GB/s (800Mhz) to 12.8GB/s (1600) and DDR4 does 29G (2133) to 45G (3xxx), most common DDR3 1333Mhz does ~ ideal 10GB/s) less memory speed. Pain in the a** to configure.

  • edited December 2015

    @jphost said:
    Thanks, you can consider it like a private cloud and run hypervisor.

    Still too vague, sorry. You want a private something that you call a cloud. People use the word "cloud" to refer to a great many different things. What are you looking to achieve? Redundancy? Automatic failover? Just load balancing/distribution? i.e., tell us things like "I want them to do things X and Y" and "if event A happens, I want it to do A and B but not C". Also, do you have the servers already? If so are they in the same place or different locations? Do you have any OS preference?

  • jphost said: @GIANT_CRAB Thank you, I was think Openstack, Cloudstack or May be Ubuntu EC (Eucalyptus powered). Do you feel any critical advantage among any ?

    Ultimately depends which you are more comfortable with. They have some different features as well. I would go with Openstack as there are many training courses offered by Redhat and Openstack foundation.

    With regards to what William has mentioned, I would also like to add on by saying that Ceph storage and Openstack can work well with each other. Redhat offers several tutorials on how to get them both to work in parallel.

    It would help you if you get certified in RHEV and Openstack, but not entirely necessary.

  • To be sure, @William describes a good setup, but I wouldn't take that as a definition of "cloud", and if you want to save costs there are lots of ways to cut corners.

    Wikipedia says cloud computing basically requires only multitenancy and dynamic provisioning.

    Thus it is possible to set up a cloud computing environment on a single server, it's mainly a question of exposing the API to launch and terminate instances.

    Thanked by 1jar
  • jarjar Patron Provider, Top Host, Veteran
    edited December 2015

    @singsing said:
    Wikipedia says cloud computing basically requires only multitenancy and dynamic provisioning.

    This. Cloud and high availability are different things, very often confused. When the two go hand in hand, it's a beautiful thing. But you can have one without the other.

    The danger of confusing these two is that you might lead people to believe that purchasing cloud servers means purchasing built in high availability.

    Thanked by 1vimalware
Sign In or Register to comment.