Stop Load Balancing VMs
-
I cannot believe how often I find small IT shops making major architectural and purchasing decisions around a desire to load balance virtual machines. There is a use case where this is important but it is quite rare. Load balancing is risky and generally expensive both in software as well as in hardware. The desire to load balance in a small business generally leads to a wealth of problems such as selecting fragile and expensive external storage where otherwise unwarranted, purchasing expensive licenses to allow for the live migration and often purchasing an extra server or two, just to justify the load balancing! This is normally quite crazy.
Load balancing in a small environment really has extremely little value. And it should be pointed out that this really only applies to on the fly and automatic load balancing which are risky operations (I've seen big firms have downtime because they did this casually, nothing is perfect.) Load balancing during downtime is easy with any hypervisor and any licensing and is just a matter of turning off the VM, moving it and turning it back on. Easy peasy. Live load balancing is not something that SMBs should be generally spending time or money on or even doing even when the option exists.
In a small environment, once loads are set, they are not going to generally fluctuate to any great degree and load balancing itself takes a large system toll. For a fraction of the cost, complexity and risk of setting up load balancing options a small shop can, instead, overprovision systems by a small amount and have more growth protection, better overall performance and no need to load balance! It's a win all the way around.
Far too often, a trivial nicety that someone desires becomes a driving force behind major decision making, often accidentally. This is one case where I often see budgets doubled to enable a feature that generally doesn't even have any value.
-
I know VMware has a function of ESXi that can do real-time load balancing and move VMs between nodes in a cluster, using similar functionality to vMotion (or possibly a derivative of vMotion) to move the machines around. However, like you said, this is truly unnecessary outside of Fortune 100 companies. Over-provision the machines or buy better machines. If some company was worried about IOPS on their disks due to a high level of SQL transactions, load-balancing in that case, in real time at least, really wouldn't be the solution. It would cost more, but maybe spending an extra $2000 or so on SSDs instead of HDDs would be more advantageous. That's just something I came up with off-hand, but I agree with your point @scottalanmiller .
-
If you are a small company and truly need this, why aren't you using something like Rackspace or Azure?
-
@thanksaj said:
If some company was worried about IOPS on their disks .....
VM load balancing doesn't even address that, CPU and memory only.
-
@Dashrender said:
If you are a small company and truly need this, why aren't you using something like Rackspace or Azure?
Cost or networking, typically. Not all workloads can go external (storage, for example.) And cloud is much more expensive than internal hosting. If your workload is not horizontally scalable they don't generally work very well for you. They meet a rather different need.
-
@scottalanmiller said:
@thanksaj said:
If some company was worried about IOPS on their disks .....
VM load balancing doesn't even address that, CPU and memory only.
Ok, then yeah, slight over-provisioning makes way more sense!
-
Most SMBs over provision heavily anyway, at least in CPU. Memory they often do. It is disks where they always cut corners and often they cut corners because of trying to do things like load balancing.
This is a case where the attempt to leverage a performance feature, load balancing, is causing them to do exactly a set of things that hurts their performance.
-
@scottalanmiller said:
Most SMBs over provision heavily anyway, at least in CPU. Memory they often do. It is disks where they always cut corners and often they cut corners because of trying to do things like load balancing.
This is a case where the attempt to leverage a performance feature, load balancing, is causing them to do exactly a set of things that hurts their performance.
I've seen some very heavily overprovisioned servers spun up by engineers...like 5GB of RAM for a Server 2008 R2 server running nothing but SW...I think it was 1 vCPU with 2 core allocated too. I argued with the engineer on that one...
-
@thanksaj said:
I've seen some very heavily overprovisioned servers spun up by engineers...like 5GB of RAM for a Server 2008 R2 server running nothing but SW...I think it was 1 vCPU with 2 core allocated too. I argued with the engineer on that one...
That's pretty light for a SW install. Normally we say a minimum of 6GB and 2 vCPU. Unless it is doing almost nothing, SW needs a lot of power to run well.
-
@scottalanmiller said:
@thanksaj said:
I've seen some very heavily overprovisioned servers spun up by engineers...like 5GB of RAM for a Server 2008 R2 server running nothing but SW...I think it was 1 vCPU with 2 core allocated too. I argued with the engineer on that one...
That's pretty light for a SW install. Normally we say a minimum of 6GB and 2 vCPU. Unless it is doing almost nothing, SW needs a lot of power to run well.
LOL I was thinking the same.. SW needs a ton of resources to run well.
-
@scottalanmiller said:
@thanksaj said:
I've seen some very heavily overprovisioned servers spun up by engineers...like 5GB of RAM for a Server 2008 R2 server running nothing but SW...I think it was 1 vCPU with 2 core allocated too. I argued with the engineer on that one...
That's pretty light for a SW install. Normally we say a minimum of 6GB and 2 vCPU. Unless it is doing almost nothing, SW needs a lot of power to run well.
Not from what I've seen. I would have provisioned no more than 3GB. The scans are mostly CPU intensive, not so much memory.
-
@thanksaj said:
@scottalanmiller said:
@thanksaj said:
I've seen some very heavily overprovisioned servers spun up by engineers...like 5GB of RAM for a Server 2008 R2 server running nothing but SW...I think it was 1 vCPU with 2 core allocated too. I argued with the engineer on that one...
That's pretty light for a SW install. Normally we say a minimum of 6GB and 2 vCPU. Unless it is doing almost nothing, SW needs a lot of power to run well.
Not from what I've seen. I would have provisioned no more than 3GB. The scans are mostly CPU intensive, not so much memory.
but really...what have you seen in your illustrious career?
-
@Hubtech said:
@thanksaj said:
@scottalanmiller said:
@thanksaj said:
I've seen some very heavily overprovisioned servers spun up by engineers...like 5GB of RAM for a Server 2008 R2 server running nothing but SW...I think it was 1 vCPU with 2 core allocated too. I argued with the engineer on that one...
That's pretty light for a SW install. Normally we say a minimum of 6GB and 2 vCPU. Unless it is doing almost nothing, SW needs a lot of power to run well.
Not from what I've seen. I would have provisioned no more than 3GB. The scans are mostly CPU intensive, not so much memory.
but really...what have you seen in your illustrious career?
I've seen plenty. I don't pretend to know it all, but every SW instance I've ever seen, it's CPU intensive on the scans but not that memory intensive.
-
I know that live load balancing is fun when you have to implement a backup solution, too. Or, at least, it sure can be. Maybe not all of the time, but every time I have come up against it it's a real nuisance.
-
@thanksaj said:
@Hubtech said:
@thanksaj said:
@scottalanmiller said:
@thanksaj said:
I've seen some very heavily overprovisioned servers spun up by engineers...like 5GB of RAM for a Server 2008 R2 server running nothing but SW...I think it was 1 vCPU with 2 core allocated too. I argued with the engineer on that one...
That's pretty light for a SW install. Normally we say a minimum of 6GB and 2 vCPU. Unless it is doing almost nothing, SW needs a lot of power to run well.
Not from what I've seen. I would have provisioned no more than 3GB. The scans are mostly CPU intensive, not so much memory.
but really...what have you seen in your illustrious career?
I've seen plenty. I don't pretend to know it all, but every SW instance I've ever seen, it's CPU intensive on the scans but not that memory intensive.
I don't pretend to know it all, either. Somehow, I just usually do.
-
@art_of_shred said:
I know that live load balancing is fun when you have to implement a backup solution, too. Or, at least, it sure can be. Maybe not all of the time, but every time I have come up against it it's a real nuisance.
If you are doing system level backups, it should work seamlessly or the load balancing is failing. But as for taking VM image backups, yeah, that's a lot more complicated for sure.
-
Yes, VM snapshots. Pain.
