If We Compute in the Cloud, We Need a Network Fabric

1Executive Summary

Before virtualization, when it came to corporate computing, companies kept their data in a box. When someone needed the data, a request was sent out over the network and routed to that box. The information was then sent back over the same path. This isn’t the most efficient way of keeping data, but since data mapped to a box, it was secure. Corporations liked that.

But as boxes became more powerful, the amount of processing power they had and work they did didn’t really take advantage of that improvement in performance. So some companies virtualized the hardware, making one box look like many virtual machines. New technologies made it easier to track what data was stored on these virtual machines, and security was maintained because the data stayed in these virtual machines. And with many virtual machines on them, the powerful servers were being used more fully. Again, corporations liked that.

But the corporations weren’t happy for long.  Even though there were many virtual machines processing data on these servers, there was still only one network. In order for each request to get to the right VM, the network had to know where that virtual machine was located. But virtualization had changed servers from single-family homes into apartment buildings. And like apartment renters, those virtual machines can move around fairly easily. When those VMs move, the network needs  the equivalent of a change of address form, but that means an IT administrator has to replace a card or reroute a cable. Requiring a person to intervene in what should be an automatic process, wastes time and costs money. Corporations hate that.

page of 3
Relevant analyst in ethernet
You must be logged in to post a comment.
4 Comments Subscribers to comment
  1. Stacey:

    Just to clarify, the Cisco UCS uses unified CNA mezzanine cards internal to the system architecture as a way of simplifying the overall system architecture. The Cisco UCS connects to the existing data center infrastructure with fully standards compliant Ethernet and Fibre Channel.

    The problem you note about swapping out needing to swap out hardware to support CNAs is a common challenge faced by customers who want to retrofit CNAs on existing servers, regardless of the vendor involved. For that reason, we expect the migration at many customer to be tied to the server refresh cycle–they will migrate to 10GbE and/or FCoE as they cycle new servers into their data centers.

    If I can offer further clarification, please feel free to ping me.

    Omar Sultan

  2. Celeste LeCompte Tuesday, July 28, 2009

    (Thanks for the heads up, Sridhar; I’ve corrected the typo.)

  3. A couple of comments:

    1. There is a typo in the company name “Emulex” (it show up as Emulux in the text).

    2. Comment on the following text:
    .”An advantage to many of these setups is that a company may be able to get away with using a smaller server because it now needs fewer ports. That would, in turn, require smaller switches.”

    Comment: While there may be fewer ports, they could be higher capacity (10 Gbps versus 1 Gbps). Same with the port capacity on the switches – one 10 Gbps port versus several 1 Gbps ports. However, it wouldn’t necessarily mean “smaller switches’, just different capacity considerations. Interestingly, the Xsigo solution requires a “fan out” – which, while reducing the adapter/port requirements on the server, might have only marginal impact on the Ethernet port requirements on the switch.


Explore Related Topics

Learn about our services or Contact us: Email / 800-292-3024