Networking towards the cloud
The quest to improve the use of IT resources is proving an arduous and complex journey.
For end users, the advent of dynamic data centers will be felt as an improved experience when using applications. For data center operators the impact will be far greater; computing costs will come down and workload automation will be possible, a prerequisite for the widespread adoption of cloud computing.
The first step along the journey occurred when enterprises started concentrating their scattered, underutilized servers within data centers. Servers were replaced with standardized hardware to reduce the server types IT staff needed to maintain, to reduce operational costs further. But the biggest change -- causing an upheaval in the data center -- is the adoption of virtualization techniques.
Virtualization enables servers and storage to be split into multiple logical versions, raising hardware utilization from 10% typically to as high as 70% to 80%. Such efficiency improvements are massively scaled given data centers can host tens of thousands of servers. Market research firm Gartner predicts that over half of all workloads will be “virtualized” by 2012.
Virtualization is also transforming data center networking, spurring standards developments to meet new requirements. These include the IEEE’s Data Center Bridging, the IETF’s Transparent Interconnection of Lots of Links (TRILL), and the IEEE’s Edge Virtual Bridging and Bridge Port Extension.
With virtualization, applications are no longer confined to single machines but are shared across multiple servers for scaling. This is changing the traffic flow within data centers. Until now the predominant traffic has been “north-south” across the three-switch-layer hierarchy of switches -- access (top-of-rack), distribution and core commonly used. With virtualization, the predominant traffic is now “east-west,” across the same tiered equipment.
The need to scale the switching while simplifying the management is also leading to new single logical layer architectures that will scale to tens of thousands of 10-Gigabit Ethernet ports. First examples of such single logical layer architectures include Juniper Networks’ Stratus and Brocade’s virtual cluster switching.
Data Center Bridging (DCB) supports the lossless requirements of storage traffic and the low-latency associated with InfiniBand. DCB promises a single consolidated network within the data center, and is being introduced as data center staffs adopt 10-Gigabit Ethernet ports.
TRILL is an important complement to DCB that enables far larger Layer 2 networks. TRILL-based networks linking switches across the data center will work without needing to turn off precious bandwidth to avoid the formation of loops. This is one shortfall of the Spanning Tree protocol.
The network must also cope with the switching of virtual machines between servers, across the data center, and between data centers, while also carrying information associated with each virtual machine for its correct configuration on the destination server.
A server’s software-based hypervisor that oversees the virtual machines comes with a virtual switch. But the industry consensus it that hardware rather than software executed on a server is best for switching. Two standards are in development to handle these virtualization requirements: 802.1Qbg Edge Virtual Bridging and 802.1Qbh Bridge Port Extension.
The 802.1Qbg camp is backed by leading switch and network interface card vendors, while 802.1Qbh is based on Cisco Systems’ VN-Tag technology. Both standards will likely be embraced within the data center.
All these networking standards are nearing completion. Yet while the protocols will soon be deployed, the expectation is that it will be at least five years and more likely 10 before their full impact will be felt.
Force10 Networks believes it will be a long and challenging transition. IBM points out how enterprises are used to working in IT silos, selecting subsystems independently, and that new work practices across divisions will be needed if the networking challenges are to be addressed. Market researcher Yankee Group points out that a lot of the future value of these various developments will be based on enabling automation, a big IT hurdle in itself.
“We all realize it is complex,” says one executive at a large service provider. “Managing pooled resources is a learning curve for everyone.”
Roy Rubenstein is the editor of Gazettabyte