Public Cloud vs. Hosted Hyperconvergence Costing Project
-
So doing cost comparisons seem to be very popular and I'm going to try to do a bit more of them. I thought that it might be useful if we had some real world workloads to use to compare the two approaches. Coming up with contrived examples is useful, but only so useful. Getting examples of what people actually need to compare would be far more interesting.
What are we comparing?
In the first corner: public cloud. Services like Amazon AWS and Vultr. An average mainline Windows server there is about $96/mo and Linux is about $40.
In the second corner: hosted hyperconvergence. We can play around with different options, but Scale and Colocation America are the easiest and are very comparable as it is enterprise, full support, single price and Tier IV datacenters with Amazon-like full time support.
Comparing these two is very useful because both are off-premises approaches that overlap in what they provide. Two different approaches to essentially identical needs for SMB customers.
Let me know some workloads and let's get to comparing!
-
Examples of what we would need:
4x VMs of 8GB, 4vCPU, 80GB each
2x VMs of 4GB, 1vCPU, 500GB each
1x VM of 32GB, 8vCPU, 160GB
3x VM of 1GB, 1vCPU, 20GB eachDoesn't have to be super exact, but closer is better.
-
On one side (HC) we are looking for a total capacity number to make sure that the pool can handle the needed capacity. On the other side we need to figure out each VM price on a VM by VM basis.
-
Also need to know if there are needs for redundancy and what it is. Redundancy is needed in different ways between the platforms. HC includes high availability built in, public cloud offerings do not. So some workloads will require two VMs on public cloud but only one on HC, for example.
-
Also the OS matters. Windows licensing has to be considered on both sides.
-
For those wondering on colo pricing, @colocationamerica is $399 for a quarter cab per month, which is super easy to calculate with. That's 10U which is enough for a firewall (1U), redundant switches (2U), a backup device (2U) and a rather significant cluster (5U of 1100, 1150, 2100 or 2150 nodes.) That's potentially a very large cluster if the 1150D style nodes are used. Far beyond the needs of a normal SMB.
If we needed more space, going to a half cab at ColocationAmerica is only a jump to $699/mo and gives us enough room to go a maxed out Scale HC3 cluster plus room to add additional storage options.
-
So for really quick comparison, four small Windows VMs on Vultr of 8GB RAM each is roughly the same cost as the colocation pricing. So the cost of the equipment is equalized from VMs beyond that point. This makes it easy to know that if you need less than those four VMs, you can't beat the public cloud in this manner. But that is one tiny company.
-
Also good to know if any individual VMs need lots of IOPS or would be happy with slow all-SATA drives (for comparing Vultr SSD vs SATA pricing.)
-
Each cost comparison should probably it's own thread, but if you post here I can fork it and add a matching title to make it simple.
-
So a really quick set of numbers....
- Starter Cluster of Three 1100 Nodes ($25K)
- Two Redundant Switches ($6K)
- NAS to use as a backup target ($4K)
- 3x Windows Datacenter Licenses ($18,000 alone)
- Enterprise Router, rackmount ($300)
Looking at 11 8GB VMs.
Result... break even on a five year cycle. Almost exactly the same cost. Except for a few things.... the hyperconverged solution is high availability while the hosted cloud option is not. And the hyperconverged solution includes backups, the hosted cloud solution does not.
There are lots of reasons that the HC solution here is the better option:
- To add backup to the hosted solution we'd need an additional $5K.
- To add HA, we would add an additional $53K, doubling the base cost!
- The HC solution has already invested in massive amounts of Windows Server licensing for unlimited VMs on the given host. We could move to more, smaller VMs or simply add additional VMs for free after this point. Additional VMs have no licensing overhead again until we add an entire additional node unlike the hosted cloud solution that continues to pay Windows licensing cost for each additional VM used.
- The HC solution has 128GB available (after HA overhead is removed) and we are using only 88GB in this example. Growing workloads by roughly 50% is available without additional cost.
- The HC solution has massively more available storage. The hosted cloud has only 1.3TB total to use.
- The HC solution is far more flexible. Instead of locking each VM to exactly 8GB of RAM, 150GB of storage we can tune each VM as needed. A VM that needs only 6.5GB of RAM need not use a full 8GB just because the steps are 4, 8 and 16GB, but can be tuned to exactly what is needed. Same with storage. This makes the potential density of VMs much higher.
To match the high availability and backup of the hyperconverged solution, the hosted cloud would cost $121K compared to the $53K of the hyperconverged cluster. At this size, only five or six VMs of the 8GB size would be necessary, if backups and HA are needed, to equal the cost of an HC cluster after five years!
But now we need to consider the colocation costs. This difference in cost would only matter if we were getting our colocation for free or "already covered" by other costs. In our example here we want to look at Tier IV (top tier, enterprise) datacenter colocation which for this is $400/mo. That's an additional $24,000 over five years.
This makes our numbers a little harder to compare. With 11 VMs, if we want backups but not HA for any workloads, the cloud computing comes out cheaper. But it is not apples to apples. If we want HA, the cloud solution is nowhere close. The final numbers are:
- Hosted Cloud Computing: $121K
- Hyperconverged in Colocation: $77K
The cloud option has more flexibility if workloads are transient. The hyperconverged option has far more flexibility for growth at essentially no additional cost.
-
Let's work from the other direction, now, what would it cost for a hosted cloud to have the same capacity as a full HC3 cluster from the example above? We already know that the Scale HC3 in colocation with full accouterments would be $77K for five years.
The limiting factor here is really RAM. The Scale HC3 has 128GB which would support 16 VMs of the example size above. That would be just over $161,000 on the hosted cloud with HA and backups.
-
Pretty much once we have breached the minimum threshold of hyperconvergence benefit, the cost of the cluster continues to improve with each additional workload that we add. Individual nodes with more RAM, more or faster storage, more or faster CPUs allow us to load up more VMs at minimal additional cost. Adding another node is a small investment, about $13K from our example above when we include the Windows Datacenter licensing, gives us the ability to run 50% more workloads for only that small additional investment. All other costs are already covered.
-
One thing that has not been included in this quick example, which is not real world at all, is support costs. Both approaches have additional costs. The hyperconverged solution has ongoing maintenance and support costs that need to be factored in. This could be just over $10K for a solution of this size. The hosted cloud solution does not have a hard support cost but does require a bit of additional time and expertise to build out and support, especially the HA aspects of it which may be quite complex and will generally require additional security, VPNs and more. This is easily more expensive than the cluster support. And not all workloads can be serviced in this way, so may not even be possible or practical.
-
C'mon people, I need some real world examples
-
Do you really need 3 nodes for your cluster example? Wouldn't that lower costs even more? or is HA just not possible without the third as a witness?
-
@Dashrender said in Public Cloud vs. Hosted Hyperconvergence Costing Project:
Do you really need 3 nodes for your cluster example? Wouldn't that lower costs even more? or is HA just not possible without the third as a witness?
Right, three nodes is the starter cluster. There is no two node cluster for this example.
-
@scottalanmiller said in Public Cloud vs. Hosted Hyperconvergence Costing Project:
@Dashrender said in Public Cloud vs. Hosted Hyperconvergence Costing Project:
Do you really need 3 nodes for your cluster example? Wouldn't that lower costs even more? or is HA just not possible without the third as a witness?
Right, three nodes is the starter cluster. There is no two node cluster for this example.
Please help me understand why the StarWinds cluster at two nodes doesn't work here?
-
@Dashrender said in Public Cloud vs. Hosted Hyperconvergence Costing Project:
@scottalanmiller said in Public Cloud vs. Hosted Hyperconvergence Costing Project:
@Dashrender said in Public Cloud vs. Hosted Hyperconvergence Costing Project:
Do you really need 3 nodes for your cluster example? Wouldn't that lower costs even more? or is HA just not possible without the third as a witness?
Right, three nodes is the starter cluster. There is no two node cluster for this example.
Please help me understand why the StarWinds cluster at two nodes doesn't work here?
I don't have numbers on that.
-
@scottalanmiller said in Public Cloud vs. Hosted Hyperconvergence Costing Project:
Result... break even on a five year cycle. Almost exactly the same cost. Except for a few things....
Yeah, you forgot the pipe to get shit out. There's $60K right there. Double it if you want redundancy, like what you would get in a hosted solution.
the hyperconverged solution is high availability while the hosted cloud option is not.
Da f***? You lose one node, you are hurting. I lose 10 nodes in a rack and wouldn't even have to get up to answer the call. And we are a small time provider. Someone like Verizon has 20 or so 32 node clusters in ONE datacenter for one customer line.
And the hyperconverged solution includes backups, the hosted cloud solution does not.
Da f***? What the hell have you been looking at? What kind of hosted provider doesn't provide backups? I just spent a cool half mil on just hardware to supply backups for folks. Verizon has 42U tape libraries backing up stuff all day long using Commvault. Not to mention the cloud backup processes if someone really didn't want all their stuff in one spot.
I think you are going in on this half baked. Yeah, you can get some stuff locally cheaper. But don't compare hosted versus local on price alone. You capped your shit out local, you are stuck until you buy more and have to sink that cost for the next cycle. I can scale your machines up with more processors on a whim and drop you back down because Boxing Day is your biggest sale day. You getting DDoS'd by China locally, you basically have to figure it out. I have massive WAFs to block that shit. You get a pimply faced youth for $12 an hour to run your shit, something blows up and he has no clue what to do. I have people with 20+ years of experience just in our support team ready to assist your dumbass who ran rm -r /.
-
How much is your cloud solution that included backups? What are the backup windows and recovery times?
When I looked at colo it included 100/100 in the price, no idea if it was redundant or not.
