nginx global setting

[dyoms]

+1 on using VMs or at least Linux Containers.
We stopped running servers directly on hardware around 2005. Since then, the servers running on physical computers are all VM hosts.

Does this mean our servers should be running VMs instead of Physical Servers? Please treat us as newborn, we dont know much I promise.

[LHammonds]

It means installing a hypervisor directly onto the physical machine (such as Proxmox or vmware ESXi). This physical machine is referred to as a "host." Then allocate resources to a virtual machine and start it up...which is a machine in a machine. Depending on the OS workload, you could run a dozen servers on a single host. Most servers will sit mostly-idle....so running multiple virtual servers on a single host allows you to make better use of the hardware. I like to run enough servers on a host to utilize 70% to 80% of the resources (I have 8 hosts). However, that also depends on the amount of hosts running. If you only have 2 servers with a similar resource configuration and want a high-availability setup, that means you need to run at no more than 40% to 45% of capacity on each host so that if one host goes down, you can run all the VMs on a single host (e.g. maintenance of host hardware). You can utilize more and more per host the more hosts you have. Such as being able to run at 60% capacity if you have 3 hosts or 70% capacity if you have 4 hosts, etc.

I started to document how to install, configure and use Proxmox but it is not "finished" documentation since I got pulled off it to work on other projects.

LHammonds

[TheFu]

People in IT have been using virtual machines or containers almost exclusively for over a decade. Some places for over 15 yrs.

There is so much added flexibility and minimal overhead in running VMs today that is just isn't worth NOT using a VM. Decoupling the OS from the hardware is a very great thing. After you've done your first migration of a VM from one system to another with nearly zero downtime, you'll become a believer too.

I would avoid all VMware stuff due to the high costs. VMware is sorta like Apple stuff. Once you buy the laptop, all those little extra devices and software also cost more. OTOH, if you say with F/LOSS ... like Proxmox, Xen-NG or KVM+virt-manager, all those little extras are free too. A quick example is backup software. On VMware ESXi, everyone uses commercial backup software. Last time I priced out those tools, they were $1000 for each system. OTOH, our KVM system use F/LOSS backup tools - source code and running as many copies as we need are $0 - and actually work much faster than the paid options for VMware.

What you get with VMware is a phone number to call. Paid training is available (but not cheap) and when something bad happens, related to the hypervisor, an expert you can pay to show up and fix stuff. That expert won't be cheap. My company worked with a certified VMware expert - he also did training for VMware. The guy retired at age 40. You don't do that being underpaid. I think his hourly rate was $200-$300. Easy access to certified people is one great thing about VMware. Everything around ESXi is expensive ... so that can be a good thing or a bad thing depending on your career goals. If you are technical, you want to know ESXi because you'll always be employed, somewhere. Those places will be used to paying for costly software AND for the people to run those systems. An ESXi admin will earn 15% more than what other hypervisor admins earn. But they tend to become silo'ed. They don't keep up with other hypervisors, so they become like MSFT-certified people - only pushing MSFT solutions. The same happens for VMware trained people.

I was one of those VMware trained people, but when I started my own company and VMware raised the costs to run their software over 50%, I started looking at other solutions. That was in 2010. Found KVM and switched all our ESX/ESXi and Xen hypervisor hosts over to KVM. I have never, not once, regretted that choice. I could have made much more money if I'd stayed in the VMware world, but my conscience would not like that knowing what I know about the other options.

Finding someone certified in Xen or Proxmox or KVM isn't as easy - though both Xen (Citrix) and Proxmox have commercial companies behind them, so you can get experts easily, for a price.

KVM experts are harder to find. That's because there isn't any formal training for KVM. Nobody is registered. Most are self-taught, like me. I'm only an expert in KVM for the things I've used. I have friends who run huge KVM server farms - over 2500 physical servers. These are part of openstack deployments. My day to day work is around a few small clients who have 3-server KVM setups, running fewer than 50 VMs.

For my home environment, I only run about 10 VMs on 3 physical systems. Most run on a single machine with the others used for testing and when HW upgrades are planned. These systems are extremely stable, BTW. If it wasn't for patching requiring a reboot, they'd run indefinitely without fail.

With all that said, I completely agree that proxmox is probably the best answer for your initial VM needs.

[LHammonds]

I completely agree that proxmox is probably the best answer for your initial VM needs.

Yes, for a Linux-newbie, it is an AWESOME introduction to virtualization. And even though I cannot claim the role of newbie on Linux anymore, I will say that I tried to setup KVM but found it a bit painful and time-consuming compared to vmware ESXi and Nutanix Acropolis...which is why I looked for other better-integrated options and found Proxmox which is very similar to the Nutanix interface which is KVM-based too...but like vmware ESXi, it costs a pretty penny.

LHammonds

[TheFu]

Yes, for a Linux-newbie, it is an AWESOME introduction to virtualization. And even though I cannot claim the role of newbie on Linux anymore, I will say that I tried to setup KVM but found it a bit painful and time-consuming compared to vmware ESXi and Nutanix Acropolis...which is why I looked for other better-integrated options and found Proxmox which is very similar to the Nutanix interface which is KVM-based too...but like vmware ESXi, it costs a pretty penny.

LHammonds

KVM doesn't mandate taking over the entire machine, unlike ESXi and Proxmox. But with flexibility comes complexity. A basic KVM setup using virt-manager really only needs a few non-KVM things setup.

• Linux bridge
• Storage Backend(s)

Manually setup Linux bridges ensure the network performance isn't poor. I don't know if this is an issue any longer, but only a NAT virtual-interface is setup by default.

Storage backends are many and extremely flexible. I use a mix, depending on the VM purpose. Most of the time, I'll use an LV. Virt-manager will "do the right thing" from allocating from a VG pool. The VM is presented with a block storage device that's very fast. The hostOS can create a snapshot of the storage which can be used to make a backup - I think that's what proxmox does too.

I don't know how clustered/replicated storage is handled for proxmox.

But KVM supports many backend storage solutions like ceph and sheepdog for when we need near real-time block replication to N+1 storage devices. openstack supports ceph too, but openstack doesn't make sense for fewer than about 20 physical servers.