Hello,
I need some advice on server virtualization. My clients generally all want “dedicated” servers, the traditional way but my cabs are getting stacked and a bit costly, cumbersome. I’m looking to virtualize my servers. Below is a sample quote from my server vendor. The below is $22k plus another $6k for 16 x 2TB SSD drives from Amazon. I use SolusVM control panel for my VPS customers which is simplified. Only downside is it cannot virtualize a bunch of physical servers as far as I know, plus I cannot add bulk IPv4 addresses in one shot when assigning them to individual VMs. Most clients lease large quantity of IPs from me.

For starters, I was looking to start off with the below sample config:

Four Intel Xeon E5-4657L v2 Twelve Core 2.4GHz 30MB 8GT/s 115W
1024GB (8 x 32GB) x (4 CPU) PC3L-10600L
PERC H710p 1GB NVWC 6G (RAID 0, 1, 5, 6, 10, 50, 60)
Small Form Factor - (SFF - 2.5in) Supports 8 Drives
Second 8-Bay Drive Cage with 16 x 2.5in Bay Backplane
16 x 2TB SSD (Micron)
Dell Broadcom 57800 - Dual 10Gb RJ-45, Dual 1Gb RJ-45 (NDC)
iDRAC 7 Enterprise with 8GB vFLASH Card
Dual 1100W Redundant Power Supply

• Do you think VMware’s vSphere Hypervisor free product is sufficient? I also considered CloudStack. Any better virtualization software you might recommend?
• If I client wants a full GigE uplink, how would that work say on above config with dual 10G and I have more than 20 clients on the same box?
• Since the above config is 48 cores total, suppose I sell 50 “virtual dedicated servers” with 24 cores each, is there a physical limit within VMware or is it somewhat like shared hosting where you can oversell bandwidth, disk space per se?
• Since I’d install 16 x 2TB SSDs (I would use RAID-0 for optimal performance and full storage amount), would these be easily hot-swappable should one drive fail? Would I be able to sell more than the physical amount allocated? In other words, if I have 10 VMs with 2TB each, that’s physically 20TB total but each VM is generally using far less. Does VMware put an actual allocation limit based on physical limitations of the hardware itself?

You have so much going on here I don't think anyone can really tell you a solid opinion.

To start off you'd need some performance metrics to see what can be performed.

Q1. There are plenty of VM solutions. Some more well integrate with your server hardware. The commercial products have features that an admin may wish to have. Some abilities are cloning online, moving seamlessly and remote access/control as well as cutting edge speed and integration. The free products may be used but metrics and features issue. XEN and KVM as well as the virtualbox or proxmox may be most common alternatives. I think ESXI may still be available?

Q2. To dedicate bandwidth you may have to search on that. I get the feeling you can assign resources and limit on some vm's.

Q3. Modern VM management tends to let resources float so to speak. You can specify directly but some admins let 48 cores be assigned to all vm's. The VM tends to select based on loads. Again you'll have to study each vm's choices.

Q4. Is kind of like cores. VM's are able to have drives that grow. The problem is if they all grow or not maintained you will run out of room. I'd consider mechanical drives to be used on some clients or some storage.

Folks still rely on hardware based raid arrays. They tend to be fast and have built in features too.
The other ways would be some sort of software or LVM/BtRFS/ZFS deal.

2 members found this post helpful.