Citrix XenServer VS Proxmox

[uberguru]

Before i was thinking i will deploy Proxmox on my colo setup but after seeing the move by Citrix to make the XenServer fully opensource..i am thinking if everyone else is thinking same...why will i want to go for Proxmox now?

I mean i am sure XenServer should prove better? Or is this a wrong way to look at it?

 

[Jeggs101]

Try both. See what you like. I would never deploy something remotely that I have not tested thoroughly.

 

[PigLover]

Both are good. They each have some weaknesses too - whether or not you run into them dknda depends on what you are doing.

For a single-node system Proxmox is great. For clustering it starts to be a PITA because of their 1980'ish view of "quorum" and the voting database they use to keep it all together. Unless half+1 or more of your nodes are up and running you can't make any changes. Fine in some environments, but for a lab where you are trying to keep most nodes off most of the time it becomes tiresome. Their HA model is also borked due to their requirement for "fencing". For a single node system Proxmox is fantastic and its pretty easy to do an AIO with ZoL running on the host OS.

Xenserver is solid, easy to manage and easy to cluster. For simple VMs you'll love it. If you do anything IO intensive, however, you'll hate it pretty quickly due to the way they process IO with everything running through Dom0, which introduces latencies and slows everything down. Adding device drivers for unusual IO is also difficult due to the special build kernel for Dom0. Finally, Dom0 is 32 bit only which creates issues with newer devices and drivers. Its also difficult to do an AIO on Xen because the IO bottlenecks make it not work well in a "guest" and the 32bit Dom0 makes it near impossible to get a high-performing file system on the host OS.

 

[uberguru]

Both are good. They each have some weaknesses too - whether or not you run into them dknda depends on what you are doing.

For a single-node system Proxmox is great. For clustering it starts to be a PITA because of their 1980'ish view of "quorum" and the voting database they use to keep it all together. Unless half+1 or more of your nodes are up and running you can't make any changes. Fine in some environments, but for a lab where you are trying to keep most nodes off most of the time it becomes tiresome. Their HA model is also borked due to their requirement for "fencing". For a single node system Proxmox is fantastic and its pretty easy to do an AIO with ZoL running on the host OS.

Xenserver is solid, easy to manage and easy to cluster. For simple VMs you'll love it. If you do anything IO intensive, however, you'll hate it pretty quickly due to the way they process IO with everything running through Dom0, which introduces latencies and slows everything down. Adding device drivers for unusual IO is also difficult due to the special build kernel for Dom0. Finally, Dom0 is 32 bit only which creates issues with newer devices and drivers. Its also difficult to do an AIO on Xen because the IO bottlenecks make it not work well in a "guest" and the 32bit Dom0 makes it near impossible to get a high-performing file system on the host OS.

One thing am not sure is if high-availability features are available for the open source xenserver....also not sure what you mean by the IO bottleneck...anywhere i can read up on this?

 

[nitrobass24]

Xenserver website would be a good place to start for HA.
Features

 

[Patrick]

One thing am not sure is if high-availability features are available for the open source xenserver....also not sure what you mean by the IO bottleneck...anywhere i can read up on this?

Thought I may posit a suggestion. A lot of this you can Google and get direct from the source in less time than typing your question. May help you get to answers to basic questions faster in the future.

 

[uberguru]

Thought I may posit a suggestion. A lot of this you can Google and get direct from the source in less time than typing your question. May help you get to answers to basic questions faster in the future.

Sure just same way we can do that most of the things we ask each other..so it comes down to getting confirmation from people and comparing theory to others real experience..just cos am asking doesnt mean i have no clue..sometimes i just want to hear other's side.

Just thought i had post that as well.

 

[uberguru]

Xenserver website would be a good place to start for HA.
Features

Don't see anywhere where there is mention of IO bottleneck there

 

[legen]

Both are good. They each have some weaknesses too - whether or not you run into them dknda depends on what you are doing.

For a single-node system Proxmox is great. For clustering it starts to be a PITA because of their 1980'ish view of "quorum" and the voting database they use to keep it all together. Unless half+1 or more of your nodes are up and running you can't make any changes. Fine in some environments, but for a lab where you are trying to keep most nodes off most of the time it becomes tiresome. Their HA model is also borked due to their requirement for "fencing". For a single node system Proxmox is fantastic and its pretty easy to do an AIO with ZoL running on the host OS.

Xenserver is solid, easy to manage and easy to cluster. For simple VMs you'll love it. If you do anything IO intensive, however, you'll hate it pretty quickly due to the way they process IO with everything running through Dom0, which introduces latencies and slows everything down. Adding device drivers for unusual IO is also difficult due to the special build kernel for Dom0. Finally, Dom0 is 32 bit only which creates issues with newer devices and drivers. Its also difficult to do an AIO on Xen because the IO bottlenecks make it not work well in a "guest" and the 32bit Dom0 makes it near impossible to get a high-performing file system on the host OS.

Does this mean that you wont really benefit from running VMs on SSDs using xenserver, is the IO-penalty that big?

 

[PigLover]

You'll still see an advantage to running VMs on SSD:

- The "IO Penalty" describes two extra context switches per system call (context switch is a transition from User space to Kernel space) caused by the way Xen forces all IO operations through its "Dom0". With modern CPUs this adds a few hundred nanoseconds per system call. While this seems small it can become materiel for large scale IO activity.

- The random IO latency of a modern spinning disk is a few milliseconds, while the random IO latency of an SSD is <1 milliseconds (often a lot less). This difference is much larger than the extra IO penalty from Xen.

- Most of the issue with spinning disks on machines with multiple VMs is that IO gets well randomized. You are limited by the IOPs of the disk subsystem much more than by raw speed. While SSDs might be 3-6x faster on raw sequential bandwidth than a spinny disk, they are often 100s of times faster on IOPs due to seek latency differences.

 

[legen]

You'll still see an advantage to running VMs on SSD:

- The "IO Penalty" describes two extra context switches per system call (context switch is a transition from User space to Kernel space) caused by the way Xen forces all IO operations through its "Dom0". With modern CPUs this adds a few hundred nanoseconds per system call. While this seems small it can become materiel for large scale IO activity.

- The random IO latency of a modern spinning disk is a few milliseconds, while the random IO latency of an SSD is <1 milliseconds (often a lot less). This difference is much larger than the extra IO penalty from Xen.

- Most of the issue with spinning disks on machines with multiple VMs is that IO gets well randomized. You are limited by the IOPs of the disk subsystem much more than by raw speed. While SSDs might be 3-6x faster on raw sequential bandwidth than a spinny disk, they are often 100s of times faster on IOPs due to seek latency differences.

Thanks for a great answer!

 

[xena]

You can take xenserver as rock solid for deployment. Theres some theories about "penaltys" there and there but in reality if I compare old xenserver 5.4 which I still have up and running few years without any problems with vmware 5.1 for example on same HW a will definitelly choose xenserver, because response from xenserver is better for me. But unfortunatelly our company "policy" dictate than we go to vmware way(management decision...even if they dont have money for that right now), and now Im force to migrate from xenserver to vmware and its pain in the ars. God I hate vmware...for every stupid little thing like HA or vmotion you must pay big amount of money. Few more years and they have more f..k up licensing model than MS and thats something . But back to you...Proxmox will be a very good choice. Actually if I want to virtualize only linux OS I will choose proxmox. For windows will be better xenserver because proxmox still using old redhat patched kernel 2.6 and theres some problems with KVM+virtio drivers+windows combo. Community even make effort and bring kernel 3 version for those interested and testers. And in whats proxmox rule badly above xenserver and vmware are default backup features.

 

[legen]

And in whats proxmox rule badly above xenserver and vmware are default backup features.

Why is it bad with build in backup ?

 

[xena]

because they does not have anything worth of word backup in default.

 

[voodooFX]

I love proxmox, even if at the moment I'm using vmware but keep in mind proxmox is not supporting USB/Serial/Parallel/etc passthrough and/or PCI/PCIe passthrough.

 

[rubylaser]

Proxmox supports PCI passthrough and USB passthrough with KVM virtual machines.

 

[voodooFX]

wow, ok
I definitely have to try that!

 

[Joe Shestak]

Personally I prefer to go with Proxmox.

 

[sasha]

Proxmox is nice

 

[kroem]

Shameless bump, 2 years on. Buy I'd be interested in knowing what the opinion is today? I'm looking to move from ESXi to a free option. (my esxi hosts are "free"...)