End of Row vs Top of Rack approaches

In situations where there big networks with many servers to connect with each other (such as in a data center), these networks must be flexible enough to allow the computing power required by large installations. Two popular network designs used in such circumstances are TOR (Top of Rack) and EOR (End of Row). We will discuss the advantages and disadvantages of both approaches in this article.
In any datacenter there are several racks of servers and storage equipment. Each rack contains multiple computing devices. The TOR approach recommends that switches should be placed on top of each rack and that all the servers of that rack should be connected to their corresponding switch. In addition, these switches will be connected to aggregation switches using one or more cables.
Among the advantages and limitations we highlight:

• Reduced cable complexity because we are minimizing the cables which exit the rack
• If the racks are small there could be a switch every three racks
• More switches are needed than in the case of EOR
• There may be several ports left unused on the switches because they are difficult to match with the servers in the rack and this means wasted ports, power and cooling
• Expanding a rack in the TOR approach can be complicated if we run out of ports. This is not very scalable

In the EOR design, each server (in individual racks) is connected to an EOR aggregation switch (also called a chassis) directly, without connecting to individual switches corresponding to each rack. Larger cables are used to connect each server to the chassis switches. There can be several EOR switches of this type in the same datacenter, one for each row or for each certain number of racks.
Among the advantages and limitations we highlight:

• Increasing the number of ports is easy since the chassis switches are designed to be able to expand their capabilities
• Chassis switches allow for better redundancy of power, cooling and control which means better availability.
• Servers can be placed freely irrespectively of how many ports we have on each switch, unlike the case of TOR. This is more practical for cooling and to use the ports to the maximum. It's cheaper in this sense.
• In the EOR approach, longer cables are needed which take up valuable space inside the datacenter, as well as making it more complex to manage and debug.
• The cost of high-performance cables required for EOR can be very high, while in TOR the high-performance cables were only the ones that connected the switches between each other. It's harder to upgrade cables in the EOR approach.
• Many servers cannot be connected by fiber yet, so that resource cannot be counted on.

In conclusion, for the design of our datacenter in the Networks Management and Planning subject, we are going to choose the Top of Rack approach since price is not a limitation, and it occupies less space, is less complex to debug and allows higher capacities. We will simply have to take into account the number of servers per rack so that we don’t run out of ports.
 
Eduard Lecha and Matías Balzamo
 
#GPXLaSalle