Today, organizations need more agility and elasticity in their computing services. Sounds cloudy, huh?

I’m not a fan of the term ‘cloud’, with the vague meanings and weird marketing tricks. We banter around at the office and come up with silly & new terms like ‘fog’ (because it isn’t clear) or ‘to the clown!’ (a play on words in reference to the Microsoft advertisements).

What I am a fan of is the change in operation for how we (all of us) use computing facilities.

Not long ago we were all buying servers for either a single use (front-end web server, a database server, etc) or a larger edition to combine services onto (perhaps a web/database server with email services planned). This is great from the hardware manufacturer point of view but isn’t the best use of the business’ cash flow, creates inflexible hardware configurations, and isn’t very green.

Enter the concept of the virtual data center (or VDC), something we rolled out on October 1st, 2011 after months of testing.

We call the brand ‘vmForge’ and the product ‘Virtual Data Center’ otherwise known as the vmForge VDC.

Companies no longer need to look at everything on a per server basis, many times over configured in one way or another. For example, customer wants to build an IIS web server, Exchange server, and a mailing list server built on a UNIX-like distribution with Mailman. Three different servers would have been purchased based on the requirements. Heading to Dell, we would end up with something like the following (which, BTW, I consider to be mid-sized components, and all rack mounted, and not including any OS licensing):

• $2,980: IIS server, 6 GiB RAM, 1 4-core processor, 4 160 GiB drives in RAID5 with a hot spare
• $4,637: Exchange server, 12 GiB RAM, 2 4-core processors, 6 146 GiB 15K RPM drives in RAID5 with a hot spare
• $2,771: UNIX-like server for Mailman, 6 GiB RAM, 1 4-core processor, 3 160 GiB drives in RAID1 with a hot spare

The above hardware is built on Dell PowerEdge R610, 1U servers with redundant power supplies and is retail pricing with whatever instant savings were available today when I configured them (October 11th, 2011). (RAM laid out in 3 DIMM configurations using 2 GiB DIMMs)

That is a total of $10,388 in capital expenditures. What’s missing? You still need to add in colocation fees per month. And there are wasted resources in each system laid out above.

The IIS server, even with a local database server, probably won’t utilize the full 6 GiB of RAM, and the disk layout is reasonable for disk I/O and redundancy but will have wasted space (most customers/companies won’t put that much stuff up publicly for download). Let’s call it 2 GiB of RAM and 200 GiB of disk space that will go unused.

The Exchange server will be happy with that much RAM and CPU seeing that it will have all roles on one system though the need for 12 GiB is kind of grasping a bit – 10 GiB would be good too but is not a valid configuration on this hardware. The disk layout is built for much higher I/O than the 7200 RPM disks in the other 2 servers but the amount of wasted disk space doubles just to give Exchange the I/O it may require.

The Mailman server is way over configured – it will never use that much CPU, things can run efficiently with 2 or 3 GiB of RAM, and at least 200 GiB of disk will be wasted even if you ran the mailing list server for 10 years with archives.

There has to be a better way…

Enter the vmForge VDC.

Build a data center with all of your resource requirements but now you can say goodbye to over-configuration of resources on a per server basis.

The number of virtual servers you want to use is only limited by the resources you have purchased. The only extra cost per virtual server is OS licensing, if required (looking at you Microsoft Windows Server 2008 R2 & Red Hat Enterprise Linux).

Build a vmForge VDC with 12Ghz of CPU, 16 GiB of RAM (3, 10, 2 respectfully), and reasonable storage requirements for each server, like so:

• IIS server, 3 GiB of RAM, 2 vCPUs, 120 GiB of storage
• Exchange server, 10 GiB of RAM, 6 vCPUs, 160 GiB of storage
• Mailman server, 2 GiB of RAM, 1 vCPU, 20 GiB of storage

And adjust each server config based on measured usage and make adjustments based on your computing needs and not by what hardware you purchased.

The above would have a cost (no OS licensing fees included) of approximately $425 per month plus bandwidth or transfer. That’s the price of colocation in a half rack with power included! Without CAPEX (or spending your hard earned cash on hardware that will be outdated faster than your financing can pay it off over 3 years).

The vmForge VDC product is powered by VMware vSphere 5, vCloud Director, vShield Edge, and Fortigate firewalls using storage based on NetApp and Nexenta.

Certifiably cooler than colocation…but that’s just my opinion.

These parts are: electrical service, backup generator, uninterruptible power supply, automatic transfer switch, power distribution units, power monitoring, and power consumers, such as servers, routers, and switches.

The main difference between power for devices in the datacenter and other types of power service is that a power loss in this situation can be a really big deal. These devices need power that is always on, even in a storm, to function properly. Making the datacenter stay on when the power from the electric company falters takes some special design considerations.

This is how the pieces fit together for our environment:

Normally, all of the power used by the datacenter comes in through one or more electrical services which are either part of the building’s power system, or are dedicated to the datacenter.

This electrical service is pretty reliable, but it can and does fail. Bad weather, natural disasters, human error, road work, and scheduled maintenance can all turn the power off for a time. In some places it is more rare than others to have an electrical outage, but no place is immune. A generator, transfer switch, and uninterruptible power supply (UPS) will keep the datacenter on while the power grid is off.

The UPS is a battery backup for the electrical service. Our datacenter uses a double-conversion, or “online” UPS. Double-conversion means that the AC power coming in from the electrical service is converted to DC power to charge the batteries, and then the DC power from the battery strings is converted back to AC power to power the datacenter. This method has some pros and cons. Since power is always supplied from the DC bus the batteries attach to, defects in the incoming power are smoothed out. During the conversion to DC and back to AC, voltage spikes and dips are corrected.

If the electrical service voltage goes too low or too high for a period of time, the load on the other side of the UPS should never see a change in the power it gets. If the electrical service drops offline for any reason, the UPS keeps providing power from its batteries without any interruption. Which is why it’s called uninterruptible.

Double-conversion comes at a price. Converting the power to and from DC, running the UPS itself, and keeping the batteries charged all take some power. For a typical UPS, this all adds a little more than 10% to the total power that goes through it.

When the electrical service goes out, the batteries power the data center forever, and they are not supposed to. This is where the transfer switch and the generator come into play. The transfer switch really does two jobs. First, it watches the grid power and tells the generator to turn on when there is a problem and to turn back off after the power stabilizes. Second, the transfer switch actually switches the feed that goes to the UPS from grid power to generator power and back when it is time.

When the backup generator gets the signal, it starts up and begins producing electricity. This usually takes a few seconds to a few dozen seconds, depending on the type of generator. Once the generator is running and the power coming from it is stable, the transfer switch connects the UPS to the generator instead of the grid. A while after the grid power comes back on and looks stable to the transfer switch, it connects the UPS back to grid power and tells the generator to shut down.

Also worth mentioning, even though it is not exactly part of the power system is the cooling. Power and cooling go hand-in-hand because all of the power that is consumed in the datacenter turns into heat which must be removed. The size and design of the cooling system is directly related to how much power the systems it will be cooling will use.

Typically, the cooling system is connected to generators, but not to a UPS. The pieces of the cooling system don’t suffer (much) from a power outage of a few seconds the way a server, firewall, or storage system would. It would be inefficient to run that extra power through the UPS when it is not needed. The lights are treated the same way. You’ll get by in the dark for a few seconds while the generator comes online. The servers are more picky.

From the UPS, getting the power to the servers should seem familiar if you are comfortable with how your home electrical service works. Here goes. From the UPS, the power is distributed by PDUs to the individual equipment racks. The PDU is a breaker panel with some additional features. Each rack gets power from one or more electrical circuits,which all connect back to breakers in the PDUs.

In our data center, each PDU also contains a transformer to convert the 480V Wye electrical service to the 120V single-phase circuits we deliver to each rack. Each of these circuits ends in an outlet that one or more rackmount power strips (also called PDUs, thank you very much) plug into. The equipment in the rack plugs into those power strips.

So in short, the transfer switch tells the generator when it is needed and connects the UPS to it when the generator is ready. The UPS smooths out the bumps so that the servers in the datacenter never notice a thing. Of course, every last part of the system is a little more complex than I have outlined here, but that’s how summaries go.

That phrase has a hidden implication trailing after it “…so what did you do?”

Well, usually, we didn’t do anything. People can accept that a car, or a furnace, or a TV can break someday out of the blue. We all know that those are machines, and we have some idea of how they work, and that they do break. The problem is, not a lot of people realize that computers are machines too. Sure, If you ask someone if a computer is a machine, they’ll say yes. However, if you show them the inside of a computer tower, they’ll usually shake their head and say “Is that it?”

It’s been clear, over my many years of tech support, that people think their computer is a magical device. And since it’s magical, it should work a certain way. Forever. And when it stops working, some mysterious outside forces conspired to break it.

My brother thinks this way. To him, a computer does what it’s supposed to, and then it stops. Then he calls me, and to him, I cast a few magic spells, prod at it chanting while waving a censor of incense in the air, and it magically stops working. They I mumble in a strange, arcane language about what went wrong. He smiles and nods, and the magic box runs again. For a while.

But when it breaks, it’s the ISPs fault, or his wife “touched something” or a computer virus caused his machine not to power on.

I tell him, it’s not any of those things, it’s him. It’s us.

Computers are beholden to humanity. They suffer from our lazy programming, our dusty houses, our animals that chew on their wires, our neurotic ticks causing our legs to kick them. Computers suffer this day in and day out while we curse them, berate them, and blame them for everything we all ultimately did or failed to do. Did install an anti-spam program. Is your computer is slow now? Did you visit customcusors.biz.co.tg? Is your hard drive always making noise? Did you forget to update your operating system? Is your email program acting erratically?

Do yourself a favor. Update your operating system. Blow out your computer’s fans and vents once and a while. Don’t install every security program ever made. Don’t install trinkets that were programmed by the lowest bidder. Treat you computer well, and it’ll treat you well.

Well, until it magically breaks one day.

ipHouse is working hard to position our network and data center for the future.  This fall, we initiated a huge virtualization project. By using high-performance, efficient, virtual servers, we have been able to reduce the total number of physical machines used for hosting email and web services without impacting performance. Mike, our CTO, has been blogging about the virtualization project in his multipart series, “Virtualization and the ISP.”

Of course, virtualization is only part of the solution. IT departments need to think about the power consumption vs. performance of every machine on their network and every machine they are thinking of getting. Decommission machines that are no longer needed. Replace outdated energy hogs and stop getting more machine than is needed for the task at hand.

For years, data center and colocation power and cooling costs have been hidden or ignored. ipHouse is working to fix that. We are looking at both the power used by our equipment and the power consumed by colocated equipment. Colocation pricing is being revised to have a power consumption component. We want customers to consider power when they are choosing servers. We are also encouraging customers to take another look at their network assumptions. Especially with virtualization, separate services do not necessarily need to be on physically separate machines. 

If you are considering a move to a more energy efficient network design and want a second opinion on server specs and choices, let us know. We have been helping customers make infrastructure decisions for more than a decade and want to make sure you make the right choices for your organization.