This week at Interop we launched (link to PR) two products key to the next stage in the physical transformation of the data center – the Summit 670T, a new high-density copper version of our successful top-of-rack product line, and a 48 port 10G copper blade for our BlackDiamond X8, the first blade of its type announced for a chassis. Now, we’ve heard a lot of talk over the last year or two about the impending move from 1G to 10G on server blades, but with the recent availability of cost-effective NICs and LAN-on-Motherboard technology from Intel, based on their ‘Romley’ platform, it is now a reality. Why is this important? Moore’s Law.
Each year servers grow in performance for a given price-point – processor speed, memory, and the number of VMs supported. We’ve witnessed the shift from 1G to multiples of 1G. The next obvious jump is to 10G. With this shift, available server bandwidth grows by a factor of two or more, while infrastructure costs, power, and the number of cables and switch ports required all decrease. Given these economics, Dell’Oro predicts that by 2015, over 90% of all servers in the data center will support 10G. This reflects a growth from less than 33% by the end of 2012. And, over 50% of the Ethernet port revenue in the coming years will be driven by the data center, partially based on this copper growth.
At the same time, data center architectures evolve, and CIOs need a choice in how they deploy switching gear. This is the difference in ‘top-of-rack’ (TOR) vs ‘end-of-row’ (EOR) designs. For TOR, the servers connect to a switch (or two for redundancy) in the rack via copper. The TOR switches then connect to the core via fiber, either 10G, 40G, or in the future, 100G. This is where the new 670T comes into play. Alternatively, all servers in a row of racks connect to a chassis (or two) at one end. These EOR switches also connect to the core via fiber. The new 10GBT blade on the BDX is ideally suited for this topology. No one way is better – choice is the watchword.