This summer, Broadcom showcased its newest solutions for cloud-scale networking for the throngs of IT professionals at VMworld. We also launched our latest solution for cloud-scale networking – the StrataXGS® Trident II.
Today, we continue the conversation around cloud-scale network architectures and how our new StrataXGS® architecture can ensure that, regardless of cloud network type and design requirements, cost-effective implementation can be achieved in volume scale.
Learn more about Broadcom’s cloud-scale networking innovations.
A critical element of cloud network scalability is the size of the forwarding tables in network switches deployed in the data center. This factor impacts many elements of data center scalability — the number of servers and the ability to load-balance and provide full cross-sectional bandwidth across switch links. In turn, these scalability elements directly impact application performance and mobility.
Traditionally, data center networks were designed with the basic premise that a server has a single identity, which was composed of one M
AC address, one IP address and a single application.
Today, virtual machines increase the density of server identities with more MAC and IP addresses, as well as numerous applications. The number and types of active addresses in the data center network (MAC, L3 host and IP multicast addresses, LPM and ARP/next-hop entries) are impacted network topology designs – both legacy and emerging. The size of each of the forwarding tables in network switches has a bearing on how cloud networks can scale. When these tables reach capacity — because the forwarding tables in switches are small — scaling problems occur.
An obvious way to deal with these challenges is to increase the forwarding table size, which also requires larger memories and high operating rates. But implementing large memory blocks on high performance silicon can be costly and consume more power. Adding external forwarding table memories is not an option either because of the significantly high throughput demanded from single chip switch implementations.
Broadcom’s StrataXGS Trident II Smart-Table technology tackles this challenge through a multi-pipeline design for performance and port density scaling, along with a shared forwarding plane. It enables maximum sizing of table resources for different data center network topologies and optimal use of available memory resources.
For example, instead of having four fixed-size tables for L2 MAC entry tables, L3 IP unicast and multicast forwarding tables, LPM routes and next-hop tables — as seen in switches available in the market today — Smart-Table enables the tables to be unified into a single shareable forwarding table.
Finally, because switch forwarding table type and size requirements vary depending on data center application and network topology variances, Smart-Table technology allows configuration of the unified forwarding table capacity into unique Smart-Table Profiles, optimized for the specific type of network deployment.
To learn more about how Broadcom approaches network virtualization and SDN using innovative network switch designs for public, private and hybrid clouds, download one of our whitepapers or follow the conversation with us via Twitter, Google+ and Facebook.
Related:
- Broadcom’s 28nm Technology: Greater performance and less power consumption is only the beginning
- Virtualization Meets the Cloud at VMworld
- Broadcom Tackles Cloud Control at VMworld
About the Author
