Currently, Ethernet standard offers an upper-limit to maximum no. of ports that can be connected on a Ethernet LAN.This limit not particular to a vendor but rather to do with standard and current electronics.
A standard Back-plane of a switch can handle several 100s of Gigabits of traffic in all. But consider this, 100 switchports -each 10GBps, we already run out of the available back-plane capacity. You can’t think of adding more switch-ports.
Also, Current Ethernet standards cannot assure FullDuplex exceeding 100GBps, reliably.
But most data centers we see today have 100,000 servers or other nodes. We segment, distribute, subnet such large network data-centers resulting in in-efficiencies and higher maintenance and manual management pertaining to more hardware.
But there is a solution. University of California (UoC) at San Diego researchers Tuesday are presenting a paper describing software that they say could make data center networks massively scalable.
The researchers say their PortLand software will enable Layer 2 data center network fabrics scalable to 100,000 ports and beyond, and they even have a prototype running at the school’s Department of Computer Science and Engineering’s Jacobs School of Engineering.
Their breakthrough is outlined in paper titled “PortLand: A scalable Fault-Tolerant Layer 2 Data Center Network Fabric” that is being presented at the SIGCOMM 2009 event in Barcelona this week.
“With PortLand, we came up with a set of algorithms and protocols that combine the best of layer 2 and layer 3 network fabrics,” said Amin Vahdat, senior author of the paper and a computer science professor at UC San Diego, in a statement. “Today, the largest data centers contain over 100,000 servers. Ideally, we would like to have the flexibility to run any application on any server while minimizing the amount of required network configuration and state.”
The data center has been a hotbed of activity among leading vendors such as Cisco, HP, Juniper and IBM, some of which are retrofitting existing offerings and others of which are building products from scratch.
PortLand will work with existing hardware and routing protocols, according to the researchers, which include graduate students who worked with Vahdat.
“Our goal is to allow data center operators to manage their network as a single fabric,” according to Vahdat, who maintains a blog called Idle Process. “We are working toward a network that administrators can think of as one massive 100,000-port switch seamlessly serving over one million virtual endpoints.”
Key to this is the development of system for servers to find one another without broadcasting their requests across an entire network. Under PortLand, switches use what are called Pseudo MAC addresses and a directory service to locate servers they need to connect, including new virtual servers. The researchers say this setup can eliminate much of the manual labor required to build a Layer 3 network.