Enabling the Next Generation of Datacentre Compute

With the increasing focus on operational costs many IT departments and data centre operators are looking at ways to reduce their energy consumption whilst at the same time increasing the compute density they can deploy within their facilities. Over the years we have seen servers change from standard rack mount offerings, a single server within a 1-4U form factor, to blades where a number of individual server blades can be mounted in larger rack moot chassis, sharing power and networking and enabling greater compute density per rack unit.  We are now seeing a new class of ultradense  microservers servers coming to the market, these present new opportunities for data centre operators whilst at the same time challenging IT to better understand the workload deployed into their data centre.

Microservers are typically a rack mount chassis that contains several 10's if not 100's of small blades each of which contains one or more CPUs that act as independent servers. The resultant system houses multiple servers in a single rack mount chassis that shares networking and power connections across all the servers. In order to achieve the server density within a single chassis small power efficient highly integrated server CPUs or SOCs are required, the Intel® Atom™ C2000 series processor fit well into this class of computing - offering full server CPU functionality - 64 bit, ECC memory, I/O, 4 integrated network interfaces all with a low power footprint. The Intel® Xeon® E3 -1200v3 processors are also good alternatives for building highly dense server offering,  although consuming more power than an Atom C2000 processor they can offer better performance per server. What's important when considering deploying this highly dense class of server is to understand the capabilities and workload that these systems can support.

There are a couple of usage models that are being considered by various industry players

Hyperscale data centre operators - these companies often need to run many small servers handling a relatively simple workload - static webpage serving or Memcached for example. For these workloads
the demand is fairly well characterised and tends to be lots of small transactions . If one server reaches capacity it is easy to issues subsequent transactions onto another server via load balancing, this works as there is no inter-relationship between successive transactions.

Facebook is seen as one of the leaders in the space with the work they have been undertaking with the Open Compute Project to define new ultradense form-factors to enable them to achieve their compute density needs. At the recent World Hosting Days in Germany Quanta demonstrated their Intel Xeon E3 & Atom C2000 based OCP solution.

Hosting providers - these companies typically provide access to physical server hardware over the internet, charging for access by way of a monthly fee. Each user has access to a dedicated server and their workload resides only on that server. Hosting companies will offer a range of compute capability to their customers

  • low end processing, such as Intel Atom C2000 processor based systems for lightweight workload, small websites, personal cloud storage etc.  An example of a dedicated hosting service based upon the Intel Atom C2000 processor is the recently announced Dedibox XC from Online.net in France, 1&1 in Germany also offer Intel Atom C2000 processor based dedicated hosting solutions, and OVH are using Intel Atom C2000 for a personal storage service.
  • Mid to high-end Intel Xeon E3 & E5 processor based systems for running business application, databases etc are available from a wide range of hosting companies.

Typically a hosting company will offer a range of compute offerings to their customers ranging from €10-15/month  to over €100/month depending on the features and compute capability of the server being

Many hosting companies will utilise industry standard rack mount servers to meet their customer's needs, but we are seeing increasing interest in using ultradense microservers to enable new service price points to be met, whilst at the same time increasing the number of customers that can be supported within a given datacentre space.

Today many businesses are evaluating the opportunities provided by the compute density that microservers can offer, for some with homogeneous lightweight workloads the benefits are relatively easy to quantify, for other with a more heterogeneous workload there is still the need to deploy a range of different server types to meet their workload demand. What's most important when considering microservers is to understand the workload being deploy, does it scale by using lots of small process or threads or does it need to scale thru increasing the performance available to an individual process. Another consideration is whether the infrastructure will be virtualised as in this case it may be that virtualising more powerful systems to support many small workloads may be a better solution than deploying many small physical servers.  There is no simple answer and every business needs to evaluate its options before making their deployment decision.

For those businesses that can use this class of systems there are a number of microserver offerings in the marketplace today from various OEMs and ODMs, these provide customers with a range of processor
choices, often with 1-4 Atom C200 per blade or a single Xeon E3 on a blade.This enables a business to elect the appropriate compute capacity for their workload - Atom for light workloads, the Xeon for more complex compute needs.