[Users] general (totally noob) questions...
While I am not completely new to running instances in KVM/Qemu, I am a total noob when it comes to clustering. While I realize that not all of these questions pertain specifically to the oVirt project, I would much rather get input from people that have experience in this environment. I am in the process of designing my next server farm and would like to know some basics that I can't seem to find the answers to via google search. Background: I have a server farm setup for a large family. It consists of 9 servers (minimal-to-medium loads for the most part, but they are in poor condition physically). My hopes are to run this same setup virtually in a 3-node cluster. I can run the entire farm in a virtual environment on a host with a dual, quad-core box with 64gig RAM. That said, I know that each of these nodes will be capable of handling the entire load by itself if absolutely necessary. For disk space, my hopes are to use two drive arrays (10 configuration?) which I can build (up and out) from without having too much trouble. Everything will be fiber, including drops to the computers and televisions. Copper will exist as well for things like POE, printers, etc. (but not in scope) Questions: 1. As I understand it, a N-to-N configuration means that there will be load balancing between the nodes as well as failover. Is the load balancing something that is manual (I have to monitor/balance the load manually, or is it done automatically?) 2. If it is done automatically, how do the loads get split up? Is the virtual machine itself the unit of load that transfers from one node to another or does it go so far as balancing services running inside those virtual machines? 3. For the fail-over, is it seemless in the sense that the user's connections don't get reset or is there a short period of down-time before the service is available again? While this isn't a big issue for me, it is something that I've been wondering about. 4. Fibre channel or FCOE? (I've spent entire evenings trying to get a straight answer through google searches, but there seem to be way too many agendas). Being that one of the virtual servers will be a media server for the televisions (new addition), my worry is lag-time (I would like to serve at least 5 televisions without lag-time). The array will also support things like file server, space for about 2-dozen www (family pages (lots of pictures)), space for mail, space for backups (rSync,Amanda), ISO boots, etc. I have the logical and physical topology drawn up for the most part, but until I understand a bit more, I'm scared _less about dropping any money into the project because I've neither seen a setup like this in action, and what I find online seems a bit vague in the way of details (dumbed-down enough for the likes of me to know if the hardware I'm buying is necessary and/or sufficient to achieve my goals). Any and all pointers, suggestions, patience, etc. would be greatly appreciated. mchauber
On 09/22/2012 08:00 AM, Michael Hauber wrote:
While I am not completely new to running instances in KVM/Qemu, I am a total noob when it comes to clustering.
While I realize that not all of these questions pertain specifically to the oVirt project, I would much rather get input from people that have experience in this environment.
I am in the process of designing my next server farm and would like to know some basics that I can't seem to find the answers to via google search.
Background:
I have a server farm setup for a large family. It consists of 9 servers (minimal-to-medium loads for the most part, but they are in poor condition physically). My hopes are to run this same setup virtually in a 3-node cluster. I can run the entire farm in a virtual environment on a host with a dual, quad-core box with 64gig RAM. That said, I know that each of these nodes will be capable of handling the entire load by itself if absolutely necessary.
For disk space, my hopes are to use two drive arrays (10 configuration?) which I can build (up and out) from without having too much trouble.
Everything will be fiber, including drops to the computers and televisions. Copper will exist as well for things like POE, printers, etc. (but not in scope)
Questions:
1. As I understand it, a N-to-N configuration means that there will be load balancing between the nodes as well as failover. Is the load balancing something that is manual (I have to monitor/balance the load manually, or is it done automatically?)
ovirt has load balancing based on cpu load for either power saving or even distribution. more types and custom scheduling are coming.
2. If it is done automatically, how do the loads get split up? Is the virtual machine itself the unit of load that transfers from one node to another or does it go so far as balancing services running inside those virtual machines?
VM is the unit of load.
3. For the fail-over, is it seemless in the sense that the user's connections don't get reset or is there a short period of down-time before the service is available again? While this isn't a big issue for me, it is something that I've been wondering about.
live migration is seamless. failover is not, since the VM/host failed. engine needs to detect it, give some grace time, then start the VM on another host.
4. Fibre channel or FCOE? (I've spent entire evenings trying to get a straight answer through google searches, but there seem to be way too many agendas). Being that one of the virtual servers will be a media server for the televisions (new addition), my worry is lag-time (I would like to serve at least 5 televisions without lag-time). The array will also support things like file server, space for about 2-dozen www (family pages (lots of pictures)), space for mail, space for backups (rSync,Amanda), ISO boots, etc.
I am obviously out of my element here for what i use for my family :) why FC and not Ethernet?
I have the logical and physical topology drawn up for the most part, but until I understand a bit more, I'm scared _less about dropping any money into the project because I've neither seen a setup like this in action, and what I find online seems a bit vague in the way of details (dumbed-down enough for the likes of me to know if the hardware I'm buying is necessary and/or sufficient to achieve my goals).
Any and all pointers, suggestions, patience, etc. would be greatly appreciated.
mchauber
_______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
On Sunday, September 23, 2012 03:13:56 PM Itamar Heim wrote:
On 09/22/2012 08:00 AM, Michael Hauber wrote:
<<snip>>
Questions:
1. As I understand it, a N-to-N configuration means that there will be load balancing between the nodes as well as failover. Is the load balancing something that is manual (I have to monitor/balance the load manually, or is it done automatically?)
ovirt has load balancing based on cpu load for either power saving or even distribution. more types and custom scheduling are coming.
2. If it is done automatically, how do the loads get split up? Is the virtual machine itself the unit of load that transfers from one node to another or does it go so far as balancing services running inside those virtual machines?
VM is the unit of load.
3. For the fail-over, is it seemless in the sense that the user's connections don't get reset or is there a short period of down-time before the service is available again? While this isn't a big issue for me, it is something that I've been wondering about.
live migration is seamless. failover is not, since the VM/host failed. engine needs to detect it, give some grace time, then start the VM on another host.
4. Fibre channel or FCOE? (I've spent entire evenings trying to get a straight answer through google searches, but there seem to be way too many agendas). Being that one of the virtual servers will be a media server for the televisions (new addition), my worry is lag-time (I would like to serve at least 5 televisions without lag-time). The array will also support things like file server, space for about 2-dozen www (family pages (lots of pictures)), space for mail, space for backups (rSync,Amanda), ISO boots, etc. I am obviously out of my element here for what i use for my family :) why FC and not Ethernet?
First, thank you for helping me understand this a bit better. Why FC... I don't know? I don't know how to figure out what kind of load it will see, so I figured that I would try for the fastest option I could afford. But like I said, some people swear by iSCSI, others by FC, and others by FCOE. Being that I've never gotten a chance to experiment with any of them (so far, I've just used a NFS/SMB server). I don't know if it helps to explain or not, but the following two links show the network I would like to set up. The first is the physical configuration. The second shows the virtual machines that the cluster will run. https://mikesplace.valleygate.net/temp/VIIgate_New_Physical.png https://mikesplace.valleygate.net/temp/VIIgate_New_Node_VMs.png Being that it's all going to be fiber and if NAS is enough, can I also assume that it's safe to consolidate the storage and the "cluster" switch and seperate them into two vlans without any kind of performance hit (don't know how busy the back-channel network would be). Thanks again for the help. mchauber
On 09/25/2012 06:10 PM, Michael Hauber wrote:
On Sunday, September 23, 2012 03:13:56 PM Itamar Heim wrote:
On 09/22/2012 08:00 AM, Michael Hauber wrote:
<<snip>>
Questions:
1. As I understand it, a N-to-N configuration means that there will be load balancing between the nodes as well as failover. Is the load balancing something that is manual (I have to monitor/balance the load manually, or is it done automatically?)
ovirt has load balancing based on cpu load for either power saving or even distribution. more types and custom scheduling are coming.
2. If it is done automatically, how do the loads get split up? Is the virtual machine itself the unit of load that transfers from one node to another or does it go so far as balancing services running inside those virtual machines?
VM is the unit of load.
3. For the fail-over, is it seemless in the sense that the user's connections don't get reset or is there a short period of down-time before the service is available again? While this isn't a big issue for me, it is something that I've been wondering about.
live migration is seamless. failover is not, since the VM/host failed. engine needs to detect it, give some grace time, then start the VM on another host.
4. Fibre channel or FCOE? (I've spent entire evenings trying to get a straight answer through google searches, but there seem to be way too many agendas). Being that one of the virtual servers will be a media server for the televisions (new addition), my worry is lag-time (I would like to serve at least 5 televisions without lag-time). The array will also support things like file server, space for about 2-dozen www (family pages (lots of pictures)), space for mail, space for backups (rSync,Amanda), ISO boots, etc. I am obviously out of my element here for what i use for my family :) why FC and not Ethernet?
First, thank you for helping me understand this a bit better.
Why FC... I don't know? I don't know how to figure out what kind of load it will see, so I figured that I would try for the fastest option I could afford. But like I said, some people swear by iSCSI, others by FC, and others by FCOE. Being that I've never gotten a chance to experiment with any of them (so far, I've just used a NFS/SMB server).
I don't know if it helps to explain or not, but the following two links show the network I would like to set up. The first is the physical configuration. The second shows the virtual machines that the cluster will run.
https://mikesplace.valleygate.net/temp/VIIgate_New_Physical.png https://mikesplace.valleygate.net/temp/VIIgate_New_Node_VMs.png
Being that it's all going to be fiber and if NAS is enough, can I also assume that it's safe to consolidate the storage and the "cluster" switch and seperate them into two vlans without any kind of performance hit (don't know how busy the back-channel network would be).
Thanks again for the help.
mchauber
NFS on 10GE would be your simplest deployment (NFS is a simpler deployment than block based storage) i'd go with simple if possible. (if 1GE isn't enough, may be for home workload)
participants (2)
-
Itamar Heim -
Michael Hauber