Colo4Hyperscale Design For Data Centers
My firm ESD is approaching 1GW of hyperscale design within 2018. In case you need to do the math, that’s 1,000 MW of data center construction, within one year alone. It’s a whole new game in the data center industry. Because of this mad rush to processing, colocation companies find themselves reinventing their business models to support hyperscale clients. The challenge is, do you overbuild to support a hyperscale market, and how do you balance your enterprise clients that require a smaller load than the hyperscale client? The answer is to adhere to a Colo4Hyperscale design model.
In the past, colocation clients had generally adopted a design model of 150 Watts per square foot (W/sq ft) of critical load, equaling a 2.5 MW block including the mechanical system support. The configuration was typically a catcher block design or a distributed redundant design. The components that supported this design were readily available for a speed to market build out, typically within a three to four month period. However, within the hyperscale industry, the design target often is 240+ W/sq ft). Additionally, the typical design capacity for a Tier 2 Network (in a distributed redundant configuration) creates a 32 MW capacity per network, which is far more than any colocation company wants to build out as inventory. The question becomes, how does a colocation company design their data centers to support hyperscale clients while economically also supporting their enterprise clients?
Hyperscale clients continue to build out their own data center campuses for their primary processing. These campuses typically, on average, equal 120 MW+ and are often reaching 400 MW in total. They are mostly built in remote areas that offer low cost per kWh, abundant fiber, and tax incentives that support data center construction. However, due to their remote locations, latency becomes an issue when servicing metropolitan/urban customers. Hence the edge compute philosophy. These capacities are far more tolerable for the colocation provider, typically ranging from 2 to 8 MW per lease. Since this capacity becomes the “sweet spot” for colocation providers supporting hyperscale, the overall design should be able to support hyperscale as well as enterprise clients.
Here are some design tips for creating a Colo4Hyperscale design.
- Reserve bus: Oversize your reserve bus from 3,000 amp to 4,000 amp input switchgear with 3 MW generators. This will require a separate electrical distribution for the mechanical equipment. Additionally, upsize the UPS system from 750 kW modules to 1 MW modules. While you will gain 25% increase in capacity, your day one cost per MW will only increase about 12%. The cost per MW difference in upsizing switchgear and generators is minimal and only impacts day one cost, and not the overall cost per MW.
In certain regions of the country where you get a better PUE, you may be able to upsize the 1 MW UPS modules to a 1.2 MW UPS module and still maintain a competitive bid among suppliers.
Space: Size all equipment to physically support above the reserve bus so that in the event that a hyperscale client reviews your drawings it can support up to 250 W/sq ft (10,000-sq-ft pod) operating at a 1.2 PUE. Building shell and core are the least costly components in a data center project, and will have a minimal impact on overall cost per MW.
Enterprise client: In the event that the colocation provider has a client that only requires 150 W/sq ft, you can swap out the fuse in the reserve bus supporting a smaller load pod. The only negative cost impact is that the suite was designed and built for a 4,000 amp input switchgear and support equipment. Again, the cost of space is minimal.
Speed to market: The previously mentioned Colo4Hyperscale design model supports equipment that is readily available with reasonable lead times. One of the key design philosophies include getting your load as close as you can to the equipment ratings to reduce cost per MW. In some cases, however, you may need to conduct level 4 commissioning at 100% of the rating while reducing your IST Cx to 90% during commissioning.
Network design: Typically, colocation providers have a dual MPOP design philosophy. In today’s hyperscale market, consider designing a Tier 2 network with 4 IDF closets (loaded at 75% capacity) to support the hyperscale tenant. In all cases, the IDF will require generator backup.
To further increase you hyperscale load, you can design at 5,000 amp input switchgear with 1.5 MW modules and a 4 MW generator. However, the lead times are not good, and the space to support this philosophy has an overall negative effect when catering to the hyperscale tenant. Our studies have found that the 4,000 amp input switchgear with 1MW modules and a 3 MW generator is the most cost effective when considering cost per MW.
The challenge for the salesperson for hyperscale is to show on paper how they can support the hyperscale tenant. The Colo4Hyperscale demonstrates supporting the hyperscale tenant without drastically overbuilding the site on day one. Additionally, the design has a positive impact on the overall lowest cost per MW for total build out.