A question about proxmox hybrid cloud to perform like Vmware vcHS

K

kumarullal

Renowned Member
Jun 17, 2009
184
0
81
LA, USA
I am trying to see if this is possible. To achieve something like vmware vcHS.
I have a 3 node cluster on my premises. I want to add 2 or 3 nodes in our data-center and connectivity through VPN.
My question is if I incorporate ceph storage with 2 x 2 TB drives (proxmox will be on seperate drive) + 64GB SSD for caching and journaling on each of my nodes (including offsite Data-center which is part of the cluster). How effective will this hybrid cloud scenario be. Will it be efficient.
So here is the scenario. I create a few VMs and run it on Local nodes. Since the nodes are using ceph storage, it should replicate the vm data on all the nodes including remote nodes. Although the initial synchronization of Data to the remote nodes will take a long time (depending how beefy the KVM vms are). Eventually, it should synchronize all the data across all the nodes.
So in the event of a disaster, my VM would automatically migrate to the data-center and should run without an issue. Am I correct in my thinking?
What is the best possible way / best practices to achieve this.
Thanks in advance
 
I think you should try a simple GlusterFS setup first. It's much easier. Gluster has option for geo-replication and it well documented. See how it goes - then try ceph.

Re your idea of having HA across the Internet: I am afraid it will be significant difference between your local and remote storage when disaster happen. However I think it still possible if you plan you storage right way. How fast you think will be your link to the remote datacenter?
 
I have never used GlusterFS. Does it work with openvz containers as well, or just KVM VMs. Do you think Ceph will have an issue?
As far as HA on a remote Datacenter, the bandwidth at our location is 20 MBPS Fiber circuit. Will that be sufficient?
Thanks in advance.
 
GlusterFS is just easier to setup for test. It can work on top of the existing filesystem which is handy if you don't have extra disk space or hardware resources for testing. Both containers and qemu images are happy to live on Gluster.

No issues with Ceph and moreover I think if you already know it and confident with Ceph just do whatever faster. With that bandwidth you can achieve what you want if not exceed a certain io. Think about what is your planned disk load and acceptable replication delay. Continuous monitoring of the replication status will really help what's going on in reality.

However there will be some work around HA. Joining all in one single cluster might be a challenge. Think if you lose connectivity to datacenter both sites lose quorum. Also fensing. and etc. etc. I can see two options: 1) Two separate proxmox clusters and manual/selfmadeautomatic vm migration between. 2) One proxmox cluster with some workaround: quorum, fensing, corosync etc.

Also you should test if corosync will happy to work across the internet/tunnel/VPN. I've seen it might be very naughty.

Everything is possible. Just think price might be too high.
 
You must log in or register to reply here.
Top Bottom