Data Center Solutions » Data Center Infrastructure

Collocation Facilities And The Myth Of The Tiered Data Center

admin — Fri, 30 Jan 2009 21:48:11 +0000

A number of years ago the ANSI/TIA-942-2005 came out as “the” Telecommunications Infrastructure Standard for Data Centers. Its purpose was to provide requirements and guidelines when designing a data center and to encourage planning for the data center earlier in the property development process.

One of the most quoted portions of the standard is the “Tiered Reliability”, which was intended as a means for determining specific data center needs required to lessen the time a data center was taken off-line. The tier breakdown assists designers with a method to quantify certain aspects and to objectively compare one center to another.

Tier I – Basic Components

single path for power and cooling with no redundant components (N)
susceptible to disruptions from both planned maintenance activities and unplanned events
may or may not have back up generator, UPS (uninterrupted power supply) or raised floor
99.671% availability – 28.8 hours (statistically) of annual downtime

Tier II – Redundant Maintainable

single path for power and cooling with redundant components (N+1)
less susceptible to planned and unplanned disruptions. Planned infrastructure maintenance may require a processing shutdown
includes generator, UPS, and raised floors
99.749% availability – 22.0 hours (statistically) of annual downtime

Tier III – Concurrently Tolerant

multiple paths for power and cooling with only one active at any time (active/passive – N+1)
planned infrastructure maintenance can be performed without disruption. Unplanned events may cause disruptions
includes raised floor, generator, UPS with each power and cooling path capable of carrying the full data center load
99.982% availability – 1.6 hours (statistically) of annual downtime

Tier IV – Fault Tolerant

multiple active power and cooling distribution paths including redundant components (2(N+1) S+S) for example each path has UPS in an N+1 configuration
can sustain planned maintenance activities and at least one unplanned event without a critical impact on load
includes raised floors, multiple generators, multiple UPS and other multiple distribution components
99.995% availability – 0.4 hours (statistically) of annual downtime

The standards went a long way to elevate the need for specific elements in the data center and the best practices for their usage. Unfortunately, once the marketing spin took off, numerous collocation vendors began touting their facilities as a Tier III or Tier IV without regard for the full spectrum of component requirements or the processes supporting those components. In addition the design standards cover other elements including:

cable and fiber designs
network distribution
rack and cabinets layout
space utilization and pathways
air flow design

Many commercial collocation facilities claiming higher fault tolerance, achieve this mark by maintaining a lower power density requirement on their tenants driving up customer costs and reducing cabinet layout efficiencies. Even when the power plant supports high electrical draws, floor load ratings may limit the cabinet’s capacity. Other times, best practices must be completely rethought, for example the best hot and cold aisle layout will fail to achieve design goals if the collocation facility locates the client in a cage space which does not support air flow management. Of course occasionally collocation customers themselves lower the reliability by failing to incorporate proper redundancy practices within their private suite or cage space.

Each business has unique up-time, cost, networking and processing density goals in mind when they establish a data center. In order to reach those goals, a balance between requirements and design standards must be struck. While the tiered data center may be an over used marketing myth, the ANSI/TIA-942-2005 provides a solid base template. This template, along with skilled engineering and management practices can go a long way to facilitate a data center build out that delivers on technical design goals at an agreed upon business price point.

Data Center Design: Keep It Covered, Bundled And Out Of Sight

admin — Thu, 25 Sep 2008 06:01:39 +0000

You only relocate a data center once (hopefully), but you manage it day after day. A big part of making your data center more manageable, dressing in all the cables and fiber. This means keeping everything covered, bundled and out of sight:

a) Cabinets have doors and sidewalls.
b) Copper cables are patched within the cabinets using the provided cable management.
c) Cables follow standard paths and do not cut across cabinets.
d) Fiber and copper cables are kept in separate trays.
e) Blanking panels will be used.

Dressing in the miles of cable and fiber required to connect all the data center equipment is an art form. With a little practice you will learn how to comb through the cables to ensure they are aligned and bundled so they stay out of the way and allow access to the hardware they connect.

A large part of a well dressed facility is having a solidly designed standardized cabinet layout along with a good structure cable design and power plan. Obviously each category of equipment will require its own layout and each should be considered and planned before any equipment or cable is ever brought into the data center.

All your designs will be a compromise between capacity, initial cost, and management. Since a majority of a data center’s TCO (Total Cost of Ownership) results from the ongoing management and maintenance, spend the time up front and perform the necessary work to keep everything covered, bundled and out of site. You will see the savings over the life of your data center.

Keeping The Lights On, In A Lights Out Data Center

admin — Mon, 15 Sep 2008 04:59:45 +0000

There are two goals in an effective data center operations organization, “keep the lights on” and continually improve operational efficiency of the data center or colocation facility. Focus on these two goals means attaining maturity rather than perfection. It also means the operational team is effectively:

Taking advantage of Moore’s Law and increasing capacity through appropriate hardware life cycle management
Receiving increased value on purchases through solid vendor management practices
Evaluating future business demand and scheduling appropriate capacity
Internally developing a service catalog along with a service centered organization
Outsourcing non-value-add or business differentiating components
Are considered an internal business partner rather than a cost center

As operational organizations improve and build an enabling enterprise infrastructure, they begin to account for both mandatory and discretionary components. Most companies recognize that components evolve from discretionary into mandatory infrastructure components. Mature operational organizations use this evolutionary time period to their advantage.

An example of this evolution is Internet Messenger (IM). IM is at a stage where many companies treat it with a try and see attitude. There may be an HR usage policy, a firewall rule or two or even a corporate IM server. However, there are few enterprises that have established an OLA (operations level agreement) equivalent to IM’s older bother, email. During this period while the service is under the service umbrella of best efforts, mature organizations develop component specific practices for Tier 1, 2 and 3. They capture metrics and build a technical capability in preparation for a components evolution.

Just as email worked its way into the business fabric, IM is becoming a standard tool in the business arsenal and as such is moving from the discretionary spending bucket to the mandatory operational service delivery catalog.

Effective operational organizations keep the lights on for mandatory components, improve overall operational efficiency and effectively evolve discretionary components.

Data Center Design: When Bandwidth Will Not Do

admin — Sun, 31 Aug 2008 11:12:44 +0000

Speed on the Internet is a measure of distance divided by time and is termed latency. “Bigger pipes” or more bandwidth, increases throughput but cannot reduce latency. Consider the example of air travel.

Is a Boeing 747 three times “faster” than a Boeing 737? They both cruise at around 500 miles per hour. The difference is that the 747 carries 500 passengers where as the 737 only carries 150. The Boeing 747 is three times bigger (more bandwidth) than the Boeing 737, not faster. And like an airplane, as long as you have fewer files (passengers) than the network can carry at any one time, increasing the bandwidth will not reduce presentation times.

From a physical perspective, in order to increase the speed at which web pages or other data are presented to customers on the Internet you need to reduce the distance. There are two strategies to reduce distance. Regional data center (RDC) or content delivery network (CDN) strategies both reduce the distance data travels to customers.

In this discussion we will look at the effects of an regional data center design architecture for reducing the time it takes data to reach customers. With an RDC strategy, you have multiple data centers (or collocation facilities) strategically placed around the globe to service those areas closest to the RDC. For this analysis we simulated an RDC in Seattle, London and Hong Kong.

In order to expose the base Internet latency a test was run sending one packet of data over the Internet and measuring its round trip time. This is referred to as a ping test. As you can see the following chart shows latency increasing over distance from Seattle.

By moving the data and servers closer to the customer and establishing a regional presence, delivery times experienced by the total customer base becomes more consistent regardless of their location. In order to simulate an regional data center presence the ping test was repeated with Asia Pacific locations measuring latency times to Hong Kong and European locations measuring latency to London. The chart below shows the results of that regional data center simulated ping test. The distances and the latency are from the initiating location to the simulated regional data center. There was a 65% improvement in the average response time utilizing the regional data center model.

Conclusion:
While big network pipes will push more data out the data center door, it will not always deliver it to Internet customers faster. In order to reduce latency, move your data and presentation services closer to your customers. Depending on application solutions or systems, significant reduction in latency can be achieved through the strategic use of regional data centers.

The tests run for this analysis accurately assess latency with for simple data requests. This investigation does not represent a specific customer experience and results will vary depending on variables such as time of day, network provider and other public network activity.