Postings
Painting the Cisco Unified Computing vision and the steps to go there…
Wanted to share my thoughts around the launch of Cisco Unified Computing System (UCS) and my thoughts around what Unified Computing means to me…
Unified Computing Vision:
server virtualization using stateless servers with virtualized adapters, riding on top of unified fabric, provisioned through a single domain of management
Virtualization today:
- Memory & GigE connectivity limiting the # of VMs a given physical host can accommodate.
- The “quick fix” approach is increasing the # of GigE ports connected to a physical host. It’s not uncommon to see a physical host with 2 x HBAs & 6+ x GigE (separating out mgmt, vmotion, prod traffic).
- Result of “quick fix” = increased connectivity infrastructure cost (ie. 48-port GigE linecard = 8 servers!) plus operational cost around managing each NIC.
- 10GE starting to make more sense, take a step further into unified fabric –> simplified cabling infrastructure & cost, 1 x CNA + 2 x CNA cabling per physical host
Virtualization challenges:
- Memory capacity still the #1 barrier
- CPU overhead associated with network processing in the soft switch when using VMware
- Legacy and “monstrous” applications that are not appropriate for virtualization, better ways to provision?
UCS takes virtualization to another level:
- jumbo memory config provisions for a larger vm footprint & accommodates monstrous applications (ie. Oracle db)
- integrated “virtualization adapter” that offloads network processing from CPU when running Vmware
- service profiles provides a mechanism to encapsulate hw state (mac, wwn, firmware, bios) and move between blades
Steps to Cisco UCS:
- customers must understand soft switching, its overhead, and the benefit of n1k
- all ESX hosts must have unified fabric. Why? Because it’s cheaper (capex & opex) than connecting multiple HBA & GigE. Even if you don’t need 10GE today.
- have production machines that still need GigE? That’s okay, N2K is here to help.
- connect w/ the server folks and think at the cpu, memory, & machine level… California is the only solution w/ minimal virtualization overhead and fully integrates the network & server layers
Comments (0)
No Comments »
No comments yet.
RSS feed for comments on this post. TrackBack URL

