25 GbE: Data Center Deployment Scenarios

By now, everyone knows about the data explosion and the great opportunities with more sophisticated real-time data analytics. Still, many companies are struggling to keep up with the new demands that have been placed upon them. To be able to store, access, compute, and analyze all that data, there must be wider pipes for the data to pass through within the network or the cloud.

Wider pipes mean greater bandwidth --  in other words, faster data throughput. With more data delivered per second, more services can be provided, more compute capabilities enabled, and more companies can use the data intelligently. However, unless the data can move into and out of data centers -- and especially within data centers -- more efficiently, bottlenecks occur and large corporations can be brought to their knees.

In the past, when data growth was more linear, it took many years for network technology to advance from one standard bandwidth to the next speed. The transition within data centers from 100 Mbps Fast Ethernet to Gigabit Ethernet took 10 years. A similar length of time transpired in the move from 1 GbE to 10 GbE. However, as the amount of data being created and used has increased, there has been demand for quicker adoption of higher speeds. For example, many more data centers upgraded from 10 GbE to 40 GbE within only a few years of their previous bandwidth boost in order to keep up with the demands of the data explosion.

There are, however, some data centers that have yet to upgrade, and there is significant pressure on them to consider their options. With the demand for bandwidth only growing, 1 GbE and even 10 GbE solutions are simply not fast enough to maintain a competitive advantage. As the amount of data continues to grow, such companies require a cost-effective option to deliver a solution that delivers higher than 10 GbE.

In 2014, a group of companies, including Mellanox, Google, Microsoft, Broadcom, and Arista, created a consortium to standardize 25 and 50 Gigabit Ethernet speeds. The rationale was to reduce the cost per bit, both in terms of CAPEX and OPEX, while reducing the infrastructure requirements. The idea is that with 25GbE, data centers could actually receive similar or better performance while decreasing the necessary hardware, thereby lowering power and cooling costs.

Let’s look at three possible scenarios of companies that are looking for more bandwidth to handle data transfers more efficiently.

Scenario 1:  A company that has been running on a 10 GbE backbone feels it must upgrade to higher speed interconnect in order to provide real-time analysis of large quantities of data. It is considering a move to 40 GbE, but is concerned about cost.

The advent of 25 GbE means that this company can now use its existing infrastructure to improve its networking. 25 GbE and 10 GbE use the same switch radix and size of interface, so the same principles of designing the network apply to each. Moreover, most 10 GbE networks use a power infrastructure that is already suitable for 25 GbE. As such, performance improves by 2.5x with only a small investment into hardware.

Scenario 2:  A company is already using 40 GbE, consisting of four 10 Gbit/s lanes, but its heavy compute demands are pressing the company to look at higher bandwidth technologies.

This company has two options to consider. It can upgrade to 50 GbE, which uses similar connectors and infrastructure to the existing 40 GbE. In so doing, the company gets a 25% increase in performance at little expense, plus an additional boost to the radix because the 50 GbE switch ASIC uses 64 ports instead of 32 ports. The other option is to upgrade the four lanes from 10 Gbit/s to 25 Gbit/s each, resulting in a 100 GbE solution.

Scenario 3:  A company that has been managing its data center on 1 GbE can no longer do so because traffic flow is congested by the large quantities of data that must now be accessed. It seems logical to the IT managers to upgrade to 10 GbE, as it requires the least financial commitment while still relieving the bottlenecks.

Whereas this was a natural thought process previously, it's no longer the only option.  25 GbE offers 2.5X performance at a similar initial cost and using similar power consumption. Of course, in all three of these scenarios, the company could also consider 40 GbE or 100 GbE solutions, depending upon their specific application needs. However, the reality is that data is going to continue to multiply at an unprecedented rate. Now is the time for any company that has a desire to grow its data center and increase its Ethernet bandwidth to evaluate both their current and future needs.