Over the past few years, a trend has emerged among many organizations, both large and small, to consider housing their computing systems in colocation (colo) facilities instead of (or in addition to) their own data centers. The reasons
Colocation pricing and power costs
Of course, since building, maintaining and operating your own data center is a costly proposition, colo pricing will play a significant part in the decision-making process. However, price alone should not be the determining factor. Moreover, when comparing vendor prices, be sure that you are being quoted for comparable facilities and any support services.
Different providers' offerings may look the same on the surface but could differ substantially in the ultimate cost of service. You can rent space by the cabinet (or even half a cabinet if you are a small firm) or in a caged-in area for multiple cabinets. There are base charges for each cabinet or per square foot for a caged area. Some colos allow you to supply your own cabinets (usually in the caged scenario) while others require you to use their cabinets.
The type and amount of power and cooling will impact the price and also the number of cabinets you may need to rent (usually based on a per-rack power limitation or watts per square foot). In particular, this comes into play when you intend to use blade servers, which typically run 4-5 kW each (some, when fully loaded, can require up to 7-8 kW). Many sites can only support 5 kW per cabinet (using regular raised-floor airflow), which means that you will be forced to rent one cabinet per blade server. This is a common practice and problem that will impact your monthly cabinet or square-foot base charges (excluding power charges). A few data centers now offer to support higher power densities, but costs vary widely once you get to 10 kW or more per cabinet. This usually involves additional floor space, a containment system, either aisle-based or rack-based, or additional close-coupled cooling systems dedicated to your area or racks.
When it comes to providing power, each provider has its own formula. Some charge per-circuit (i.e., a fixed price for outlets at 120 V/20 A or a 208 V/20 A), while other sites may provide metered usage-based charges. The number of circuits you need and how heavily you load them will greatly impact your monthly costs. For example, if you are in a fixed-price, per-outlet contract and your equipment is only drawing 9 A from a 20 A circuit, then you will be paying the same price whether you draw 1 A or 16 A (please note that under the 80% rule for branch circuit protection, you can only safely draw 16 amps). Not all providers have fully metered branch circuit monitoring, if they do, they are under no obligation to charge you for actual power used. You may find that each provider has different terms in their contracts that allow them to "audit" what equipment you have installed in your cabinets and how much power you are actually using.
In many cases it may pay for you to know how much power your equipment will actually draw under normal conditions (as well as the maximum power under peak loads and at startup) before you install it in the colo. The purchase of a pair of metered rack power strips can pay for itself many times over. This will allow you to know beforehand how much equipment you can safely load per circuit and maximize the use of the circuits that you are paying for. If you are planning for high-density computing, think in terms of 208 or 230 V power, and also consider three phase power to allow for future growth flexibility and energy efficiency.
In addition, you will presumably want to have A-B redundant power feeds for your IT equipment. The prices vary for a secondary or redundant circuit, typically ranging from 20-50% of the price of the primary circuit (plus a setup or installation charge). However, remember that even though your IT equipment's dual power supplies will normally load share, the total of the A-B circuits can not exceed the maximum allowable draw of the primary, or "A" circuit. This is an extremely critical yet often overlooked rule to prevent cascade failure. Also, the colo operator will not want you to use more power than you are paying for. Colocation providers will usually have this as part of the contract, and they have the right to do a power audit to ensure that you are complying.
Colocation connectivity considerations
Now that you have dealt with the power and cooling issue, you will need connectivity to carriers. Some colo sites are carrier-neutral and allow you to connect to any carrier while others only allow you to connect to carriers that they have in-house relationships with. The costs of bandwidth and the choice of carriers and carrier diversity needs to be considered carefully. How is your connectivity delivered (i.e., the handoff and demark point). In some cases, there are cross-connect charges in addition to the basic bandwidth and port access charges. Are you getting Internet connectivity from the colo's network via Ethernet and billed as part of a metered bandwidth and demand contract? If your connections are directly to the carriers, what are your connectively options and bandwidth upgrade path: T-1, (with N x T-1 steps), T-3 (or fractional), OC-3, OC-12 and any contractual clauses, such as the term length to coincide with your tenancy in the colo site. Also be aware of who you are contracting with for your services -- the carrier directly or the colo operator? This can affect your future options, should you decide to move to another colo or perhaps change your operational model back to your own data center.
Consider the timing of the transition to a colo. An important factor to consider is the age of your existing IT equipment and its expected life cycle. If your gear is 2-3 years old and nearly at end-of-life, it usually does not pay to move the equipment. Consider planning to purchase (or lease) and install new equipment at your colo. There are several advantages to this: The first is that you can continue to operate your old equipment until you are ready to cut over to the new systems. This will allow you to install, test and troubleshoot your new installation offline, without any impact on your existing production systems. Also, the new IT gear is more energy efficient and will presumably lower your monthly power requirements, which should lower your monthly power costs. Ideally, the length of your colo contract should coincide with the expected lifecycle of your IT gear (i.e., three to four years). By starting your colo contract at the same time as your new equipment is installed, you will be in a better position at the end of your contract to renegotiate your renewal with the colo operator and then refresh your IT equipment or be free to move to another colo and purchase new equipment to install there.
Redundancy, availability and service levels
As far as the reliability of their data center, be aware that availability numbers and claims are usually based on generalized interpretations of the tier system (1-4), either based on the Uptime Institute or the TIA 968 standard. There are many tier 1 and 2 sites that have very little or no downtime issues and some tier 3 or 4 sites that have had problems. Having N+1 or 2N power and cooling redundancy is either a mandatory requirement or simply "a nice to have," depending on your business requirement and budgets. While it likely that a site with more redundancy will reduce your chances of experiencing a problem, clearly you should review the site's operational history. Moreover, examine the provider's contingency plans and recovery procedures (as well as your own) in the event a problem. The best advice is to ask for customer references that have equipment for a year or more at the particular site that you are considering.
Examine the contractual language of any service-level agreements (SLAs) to see what levels of availability and remedies are promised. You may find that if there are problems, you may have limited options or methods of recourse to break the contract, depending on the terms of your SLA and exclusions. It will not mean much to you if you get a one-day credit or even a month's credit, because you were down for "more than an hour" that month, the impact to your business will far exceed the value of the credit.
Does the promised level of availability exclude any planned downtime for maintenance? It is easy to say that there are "five 9s" if you exclude "planned" downtime. There is no doubt that most colo operators work very diligently to ensure that you do not experience problems. Nonetheless, sometimes the unplanned happens and the procedures and resources they have to mitigate the problem and restore normal service is what differentiates colo operators.
Regarding availability and continuous operation, how much fuel does the colo have stored locally (i.e., 24 hours; three days a week)? During a small localized utility failure (hopefully they have two feeds from different sub-stations), 24 hours of fuel may be fine, but three to seven days offers a better safety margin. During an extended widespread crisis, the relied-upon expectations of daily refueling may prove to be difficult, if not impossible, to achieve (case in point, Hurricane Katrina). In the end, you will typically pay more for the colo with the greatest levels of redundancy, resources and better SLAs, but it's foolish to assume that nothing will ever happen to impact the operation of your computing systems because you are in a highly rated site.
Maintenance and support with a colo
Also review the provider's overall security and access procedures as well as any hands-on support options it offers. Not all colocations have 24-hour on-site staffing. Depending on whether you intend to do your own equipment installation and support, you will want to be able to have the third-party support of a manufacturer's field engineer to gain access to your equipment at any time (with your proper authorization) to repair a server without your making a trip to meet the engineer at the site. This is particularly true if the colo site (or sites) is far removed from your main office. In addition, does the provider have equipment receiving and staging areas? Will it receive and accept packages and store them for you (and for how long -- a day? a week?). Does the site offer the convenience of a staging area to de-box assemble, test and configure any new equipment before final installation in your racks?
Some sites will not let third-party vendors enter the data center unless they are met and accompanied by a person from the customer that rented the rack or cage. This is a good security practice but may prove to be burdensome if you are not conveniently located near the site. For generalized security purposes, the location of the data center should be far enough away from most recognized potential threats or high-risk targets such as airports, major highways or railroads, yet still be easily accessible to your IT support personnel. This may be a challenge depending on the extent of local transportation system and whether your IT personnel can or must drive to get to the data center.
Clearly, if you have chosen a site for geographic diversity that is several hundred (or thousand) miles from your offices, you need to understand which support services are offered (i.e., 24-hour on-site techs) and the cost of those services. This can include providing such routing tasks as changing tapes and preparing them for pickup or shipment to your tape storage provider. Or in some cases, providers may offer different levels of support with different skill sets -- unmanaged, per-incident or hourly rates or in blocks of time per month. They may also offer a fixed monthly fee for fully managed support services. Do they utilize outside personnel, or in-house expertise, trained and proficient in supporting particular major brands of equipment such Cisco, Dell, EMC, HP, IBM, etc.
Remember, while the benefit of moving into a colocation site is to reduce your infrastructure capital costs and maintenance responsibilities, it requires that you understand that it does significantly change your existing operational model. You will not be able to just step out of your office and walk down the hall into your own data center. Keep in mind that you can no longer just re-boot that problematic server that usually hangs once or twice a week just by opening the rack door and hitting the power button.
If this is your first experience with a colo, consider practicing in your own data center first. See what you need to do and what equipment and software you may need to install to "remotely" manage your own data center while it still only down the hall. Once completely understand all the simple tasks that you or your staff do (and take for granted) because your IT equipment is in the next room, you will be better prepared to move into a colo environment. If you are not fully experienced at migrating and operating in a hands-off remote management environment, you may want to consider engaging a consultant to help you evaluate what is required. This would guide and assist you in analyzing the true scope of work and the projected total costs (in both capital investment and for any recurring support costs). They should also be able to help you evaluate the different colo vendor offerings and contracts. Ultimately, if you so choose, the consultant may also provide project management assistance and contingency planning and support during your transition to the new site.
The entire colo market is going through a period of huge growth and the larger players are acquiring smaller operators or merging to gain footprint or market share. You should also consider the size of your organization and its requirements as well as your potential importance to the provider as a customer. Get to know who you will be dealing with on a regular basis, not just the "account executive" that you may never see again once you sign the contract. Will you be a significant customer to a smaller colo and hopefully get great customer service (assuming, of course, that they have the resources) or will you just be a small fish with two cabinets in a large 50,000-square-foot "pond"?
A final thought: Moving your organization's computing systems into a colo environment is like getting married. Shopping for a colo is like looking for a mail-order bride -- there is no real dating period. Once you make your final selection you are instantly wed (sans honeymoon) and immediately move in a new house (at least for the length of the contract). And just as they say at the altar, it is "for better or for worse," so choose your partner carefully.
About the author:
Julius Neudorfer has been CTO and a founding principal of NAAT since its inception in 1987. He has designed and managed communications and data systems projects for both commercial clients and government customers.
This was first published in May 2010