I am starting a discussion to serve as a container for all the questions I am going to have along the way to building my ideal virtualization platform. Throughout the discussion I will likely ask some novice and bizarre questions. To give you a little background about myself:
I am a seasoned IT generalist / consulting engineer and architect. I've been building virtual platforms for over 8 years. I have 7 full HA platforms 'under my belt' at this time. Until now, all 7 pools have been built using Citrix XenServer. I am deeply familiar with storage systems. I have implemented multi-array Dell Equallogic SAN gear in a few of those 7 pools. If i had to say what my strongest expertise is, i would say Networking.
My goal here is to roll my hardware into a co-lo facility, hook it up to the internet and to start selling services to customers who want fast and reliable virtual machine hosting that provides private VLANs, HA migration between nodes, and SSD-like file access speeds. I want to be able to sell virtual machines anywhere from 512MB RAM with 10GB storage, up to 64GB RAM with multi-terabyte data stores.
I've already put alot of thought into this but I'm starting the discussion here and now to help me work through the details.
Here's my general game plan:
*Figure out what software configuration will work the best for my needs
*Figure out what hardware is available and select the optimal configuration for my desired sizing.
*Create a final list of all the hardware and software licensing costs
*Create diagrams and written business plan and take it to the bank (or maybe kick-starter)?
*Buy all the hardware, drop shipped to the co-lo
*Provide build-out diagrams to the remote hands
*Configure and test/benchmark the platform, fine tune for performance
*Test HA functions, Backup systems, Maintenance Activities
*Configure and deploy "Hyper-visor Automation" components, to facilitate a web interface that tenants can use to purchase, manage, migrate, and deactivate their virtual machines, to be integrated into payment and billing systems
*Make money
*Pay Loans
*Expand the pool (add additional hyper-visor and storage nodes)
*Retire Rich or have a heart attack before getting there.
I expect this to be a long term discussion (I hope).
I think Hyper-V with Load Balanced Fail-over Storage Clusters might be the way to go. I've looked at OpenFiler and OpenE server which both provide iSCSI services in HA mode however they are both thousands of dollars per year per node and only support 1-4 nodes. I expect as the system grows I will be adding 2 storage nodes at a time so this will not work.
An initial system might look like 4 nodes... 2 are hyper-v nodes and 2 are storage nodes. They would be connected over a minimum of 10Gb Ethernet. The storage nodes would be populated with 6 drives in RAID 50. The hyper-v nodes would be populated with 6 SSD drives in RAID 0. Ideally, all nodes (storage and hyper-v) would be disk-less boot.
So down to the details of where to start the Q&A... With windows on the storage nodes, can i present a single large pool of storage that is is load balanced and fault tolerant up to and including the loss of a whole storage node?
Q2: Can I do something on the HyperV node that will allow the VMs to have access to that shared storage via a local SSD cache? IE: providing lightning fast local storage speeds to the VM when writing data and re-reading cached data?
w.r.t Q2: In XenServer world I would typically do this by pointing the hyper-visor to use 127.0.0.1 as it's iSCSI target then using an iSCSI service i would present access to an SSD cached folder (dm-cache) that is backed up by an iSCSI connection to the actual iSCSI shared storage target. so Q3, is can I do the same with Hyper-V and Windows? If so, what roles, tools, services would I expect to install to support these?
Initial question number 4: Any general thoughts, concerns or gotchas around anything I've described thus far?
I don't expect anyone to answer all my questions, but if you know any part or even if you contribute nothing, thanks in advance!