Is the Time for VMware in the SMB Over?
-
@scottalanmiller said:
As much as I love moving away from on-premises servers, I don't believe that the era is nearly over. Here is how I feel, without putting a ton of thought into this question:
- While most workloads in the SMB should not be on-premises, about 10% should be and will remain this way for a long time, slowly lowering but never completely going away. Workloads that will long remain will be heavily cache and security based (proxy servers, DNS, AD, scanning, filtering, machine controllers, etc.)
- While most workloads should not be on-premises, many will remain because SMB folk have slow upgrade cycles and tend to wear tin foil hats (both IT and managers outside of IT) and do not apply logic and business acumen to these decisions in many cases. So on-premises workloads will exist where they should not for a very long time.
Aren't colos often much more expensive for SMB than onsite? Granted, the Colos often provide much better services, the SMBs have been getting away without those features in their locally hosted solutions for 2+ decades now, why the sudden need to change?
Additionally, the network performance of onsite at 100 or 1000 Mb is considerable compared to running everything from the cloud (colo).I want to accept the desire to move to the cloud/colo but I can't see how it doesn't drastically increase costs, and possibly drastically affect performance (bandwidth).
-
You can get unmetered lines to many colos. And for example our fiber lines at the county allowed us to have some stuff unmetered, like if we got a colo that was using them directly or hurricane electric.
-
@Dashrender said:
Aren't colos often much more expensive for SMB than onsite?
We found even ten years ago that the cost of power and cooling alone paid for a top, enterprise colo with 24x7 service and big time networking and redundant everything. We've found that there is no way to cost justify on on-premises server under normal conditions because colos are so much cheaper.
There are special cases when this is not the case, like when the on-premises cannot get reasonable network connections. But in general, colo is the cost savings option on top of all of the normal benefits.
-
@Dashrender said:
Granted, the Colos often provide much better services, the SMBs have been getting away without those features in their locally hosted solutions for 2+ decades now, why the sudden need to change?
Couple reasons:
- Cost savings
- Getting away with things doesn't mean that that should continue. Getting away with paying too much or not having adequate insurance or not having what is ideal is one thing, but every business should always attempt to do what is best for it. Just because you don't have to be perfect to succeed doesn't imply that you won't succeed better doing what is better.
-
@Dashrender said:
I want to accept the desire to move to the cloud/colo but I can't see how it doesn't drastically increase costs, and possibly drastically affect performance (bandwidth).
Not sure where you are seeing the cost coming from. How much do you think putting a server into a colo costs?
-
Also remember that the cost of a server in a colo is lower. Servers last longer. Their parts just don't wear out as fast when the temperature is stable and there is low dust. You can replicate this onsite but it is hard to get the quality HVAC, electric and vibration reduction of an enterprise colo. This is why colos often cut downtime by more than half. If you expect an outage from a normal server on-premises every six years, a colo might make this same hardware average go to more like ten to twelve years.
There is a reason why we see six nines of reliability from non-redundant servers. A quality setup really saves a lot of money!
-
@thecreativeone91 said:
You can get unmetered lines to many colos. And for example our fiber lines at the county allowed us to have some stuff unmetered, like if we got a colo that was using them directly or hurricane electric.
Yes, when you are small (say 2-3 servers) using open Internet is normally cheap and adequate. Most colos that we work with give us between 100Mb/s and 1,000Mb/s per server included.
When you get bigger getting a dedicated, super high speed pipe directly to the colo is generally easy and while not cheap, is easily offset by other costs and can give you LAN-like performance even to a remote facility.
-
Colos do more than just these things too. Moving to a colo also means that you are changing your network design and certain things start to happen automagically by the nature of the change in planning....
- You tend to start securing your environment for more general use cases rather than specific ones improving flexibility.
- You tend to be prepared for site failover and work from home options.
- You are well prepared for mutiple sites.
- You tend to focus on sprawl reduction and often trim expenses through better planning.
- You tend to get network upgrades and other features over time, inclusive.
- You get the HVAC, power conditioning, generators, monitoring, 24x7 support staff and other features that you "should" have on-premises but either pay far too much for, do without or attempt but likely cannot do as well.
It is mostly "tends" to be these things. And you can do much of this without a colo, but it would be far more costly and very tempting at some point to skimp on them. You have to not only look at the ongoing costs of on-premises cooling, power and monitoring but also the cost of the extra outages that you are protected against or can be, if you chose to be.
-
@scottalanmiller said:
@Dashrender said:
I want to accept the desire to move to the cloud/colo but I can't see how it doesn't drastically increase costs, and possibly drastically affect performance (bandwidth).
Not sure where you are seeing the cost coming from. How much do you think putting a server into a colo costs?
The last time I saw a quote it was like $100/month per U or more. I currently have 15 U, that would $1500/month, $18K a year. Seems difficult to get over the price tag.
Also, what about the additional bandwidth needed to talk to that datacenter? If I put my AD boxes there, I figure I'll need at least 10/10 bandwidth to ensure a reasonable experience (and that might be low). Assuming I pull all of that over the same ISP line and have no downtime (yeah right on standard business connections). Unless the price is significantly lower, more like $2-5/month/U I'm not sure I see the savings?
-
@scottalanmiller said:
...t also the cost of the extra outages that you are protected against or can be, if you chose to be.
I'm less worried about the outages at the DC than I am at my own site.
For example, to ensure as little downtime as possible we have a 10/10 Mb fiber connection (dual rings - though only one route into the building). In 8 years we've only had 13 mins of non scheduled downtime, and those mins came in two blips last spring. The bad thing is we pay $860/month for this.
On the other hand our other three locations in town all have HFC. All of those locations have had multiple outages over the past 8 years, one of them lasting more than a day. On average I'd say an outage lasts approx 2 hours when it happens on HFC.
We have new options available in town now so I'm considering dumping our Fiber connection and moving to two different ISP connections into my main location and setting up alternative routes at the firewall to cover outages. Moving to dual 50/10 connections will cost me around $500/month versus the $860 I pay now, and will increase my download by more than 5 times (10 if I can use both pipes at the same time)
-
@Dashrender said:
The last time I saw a quote it was like $100/month per U or more. I currently have 15 U, that would $1500/month, $18K a year. Seems difficult to get over the price tag.
Well anyone can find inflated pricing. But that is super misleading. When buying by the U you are paying "per server". That not $100/U. That is $100/1U Server, I suspect.
We get $50/1U server and like $80 for a 2U server. See how it is not "per U"?
Once you go beyond "per server" pricing you get things like 10U, quarter rack, half rack and full rack pricing which is far, far cheaper "per U" than you pay if you pay per each server being racked.
So a half rack might be $450/mo.
And you missed by point about improving your planning. When you run on-premises you deal with "U sprawl" because the space is free and everything else is not. In a datacenter, the space is what is not free. So with a change in planning, I bet you could reduce that to 4U - 6U if you were planning for space as a concern.
What do you have in that 15U?
-
@Dashrender said:
If I put my AD boxes there, I figure I'll need at least 10/10 bandwidth to ensure a reasonable experience (and that might be low).
Is that for a few thousand users? AD uses effectively nothing. A T1 can support hundreds of users no problem. A T1 is 1.544Mb/s. And you don't need symmetric for AD either.
-
@Dashrender said:
Unless the price is significantly lower, more like $2-5/month/U I'm not sure I see the savings?
Would you see no benefits at all from having a good WAN connection? What do you have now? Is it not needed to be reliable?
How much are you paying for HVAC and power today? We found that these costs alone paid for the colo.
-
@Dashrender said:
For example, to ensure as little downtime as possible we have a 10/10 Mb fiber connection (dual rings - though only one route into the building).
Wait, I thought that you were using the need for 10Mb/s WAN link as a reason that you can't afford a colo. But if you already have that, that's a sunk cost and can't be included here. No additional WAN cost to go to colo, right?
-
@Dashrender said:
We have new options available in town now so I'm considering dumping our Fiber connection and moving to two different ISP connections into my main location and setting up alternative routes at the firewall to cover outages. Moving to dual 50/10 connections will cost me around $500/month versus the $860 I pay now, and will increase my download by more than 5 times (10 if I can use both pipes at the same time)
So the colo just gets better and better as an option just based on general good business practices. You save money, move to something more modern and the colo just gets more and more performant as part of the normal shift in technology.
A colo is not a pure win, but everything you are describing suggests that likely you are a very good candidate for at least part of your workload to go to a colo.
What workloads besides AD do you need? AD is the most trivial, pretty much, of any possible workload.
-
Some pricing for comparison....
At the enterprise datacenters that we use...
10U quarter cabinet is $275/mo
22U half cabinet is $475/moThat means either you can add up to 7U and still be under 33% of the cost you were assuming. Or you could trim the fat down to 10U (which should be easy for an SMB, nearly all can run in 4U or 6U at a stretch if well planned) and drop to around 15-20% of the cost that you were thinking.
-
And if you could drop to closer to 4U, you could probably get down to around $200/mo.
Don't forget that colo is not all or nothing. You can move strategically the workloads that make sense to a colo and not the ones that don't. So often moving a single 1U or 2U server is what makes sense and is generally $100/mo or less.
-
@scottalanmiller said:
Well anyone can find inflated pricing. But that is super misleading. When buying by the U you are paying "per server". That not $100/U. That is $100/1U Server, I suspect.
I didn't think it included the server, but perhaps it did and I didn't understand.
And you missed by point about improving your planning. When you run on-premises you deal with "U sprawl" because the space is free and everything else is not. In a datacenter, the space is what is not free. So with a change in planning, I bet you could reduce that to 4U - 6U if you were planning for space as a concern.
What do you have in that 15U?
Oh we definitely have sprawl right now because we have an old tower style P Series converted to rack mount (with luck gone in 1 year - but I've been saying that for a year already), we have another old 3 U IBM that will be docomp'ed this year, two 2-U VMWare servers and two - 2U physical servers (could be virtualized down to a single machine but would require more storage) for our old EHR product, keep for lookups.
-
@scottalanmiller said:
@Dashrender said:
If I put my AD boxes there, I figure I'll need at least 10/10 bandwidth to ensure a reasonable experience (and that might be low).
Is that for a few thousand users? AD uses effectively nothing. A T1 can support hundreds of users no problem. A T1 is 1.544Mb/s. And you don't need symmetric for AD either.
That's more for the file and print serves than AD.
-
@scottalanmiller said:
@Dashrender said:
For example, to ensure as little downtime as possible we have a 10/10 Mb fiber connection (dual rings - though only one route into the building).
Wait, I thought that you were using the need for 10Mb/s WAN link as a reason that you can't afford a colo. But if you already have that, that's a sunk cost and can't be included here. No additional WAN cost to go to colo, right?
That's not right - the 10Mb we have now barely supports our VPNs and Cloud EHR. There is a noticeable drag when someone is downloading large files from the internet.
