Experts are predicting that the world may soon face a severe network bandwidth shortage, either through capacity not growing fast enough or through problems like the recent undersea cable failures in the Middle East. However, new optical network standards aim to address these concerns and provide increased capacity for both the wide-area network backbones that make up the internet and the high-speed plumbing that carries traffic around inside datacentres.

One firm aiming to bump up carrying capacity is optical network specialist Infinera, which is developing a photonic integrated circuit (PIC) said to be capable of delivering aggregate data rates of 400Gbit/s along existing optical-fibre network connections.

The PIC combines 10 40Gbit/s streams using so-called differential quadrature phase-shift keying (DQPSK) modulation. It integrates more than 100 optical devices on a single chip, according to Infinera, more than double that of today’s 100 Gbit/s PIC devices.

PICs are analogous to the electrical circuits used in standard copper network interface cards (NICs), but data is moved by electromagnetic radiation in the visible and far infrared spectrum, whereas data over copper is moved using different electrical voltages.

Infinera has also demonstrated data transfer speeds of 100Gbit/s over a 320km optical fibre, with a PIC using on-chip amplifiers at a wavelength of 1490 nanometres in the S-band ­ a region of the electromagnetic spectrum not typically used for commercial data transfer.

According to Infinera, standard Dense Wavelength Division Multiplexing (DWDM) optical networks are limited to using a specific range of the optical spectrum. This limitation is imposed by Erbium-Doped Fibre Amplifier (EDFA) technology used in standard DWDM that only works effectively over around eight per cent of the 55THz-wide frequency band available for use by optical-fibre networks.

Infinera optical systems director Steve Grubb said his company’s PICs offer the possibility of using optical networks across the full fibre spectrum. “As we look for ways to economically scale network capacity to respond to increased demand from IP traffic, this significantly expands the current DWDM tool kit,” explained Grubb.

But it is not simply the long-haul pipes of the internet backbone infrastructures that are starting to reach saturation point. With the growth in Web 2.0 applications, and global corporations increasingly moving data into centralised data farms, telcos and internet service providers (ISPs) may soon need to deploy such technology sooner rather than later.

Server virtualisation
Demand for increased network bandwidth is not just coming from ISPs; large enterprises are also beginning to feel the pinch in terms of datacentre bandwidth. One reason for this is the boom in server consolidation using virtualisation, which means each physical server is capable of handling a larger volume of data. The industry is tackling this problem by offering faster Ethernet kit and I/O virtualisation, which enables some of the network handling to be offloaded from the hypervisor to the NIC.

At VMware’s recent European VMworld event, network I/O virtualisation was one of the key topics. This is the process of assigning a defined amount of bandwidth for virtualised applications by segmenting the NIC into hardware-defined network paths.

At the event, Neterion vice president Ravi Chalaka said more and more enterprises are being attracted to server virtualisation as a way of consolidating hardware and minimising power requirements, but that this is putting a strain on connections between servers and storage. He added that server virtualisation “would give enterprises added incentives to move to 10GbE fibre connections from the current Gigabit Ethernet connections”.

Neterion, which sells 10GbE kit, showcased what it said was the first 10GbE products supporting virtualised I/O, the X3110 Xframe V-NIC series adapters. These can virtualise 17 bidirectional network paths for applications running under VMware’s ESX Server platform, by presenting each virtualised application with its own independent physical path using separate direct memory access and interrupts.

The adapters basically pretend to be multiple NICs, and also offload some of the virtualisation work from the hypervisor, a process which Chalaka said gives the Neterion adapter “a four times increase in I/O performance, a 50 per cent reduction in CPU utilisation and can increase application density by between two and four times”.

Another benefit of using 10GbE adapters rather than the Gigabit Ethernet equivalent, apart from the simpler cabling and management, is the power saving. Neterion calculates that one of its 10GbE adapters uses 15W compared with around 80W for 10GbE links, equivalent to 3.25Kw when spread over 50 servers.

The Neterion Xframe V-NIC is being developed in conjunction with the Single Root I/O Virtualisation standard (SR-IOV), an extension to the PCI Express standard. SR-IOV is a standard for making hardware I/O adapters, which have physically independent I/O network paths actually in the chipset.

Another option for firms looking to beef up their datacentre connectivity is the 10GBase-T 10GbE over copper standard. Cabling for 10GBase-T has been available since the standard was ratified in summer 2006, but uptake has been held back by a lack of switches with 10GBase-T ports.

This hesitancy on the part of the switch makers is due in part to the relatively high power requirements of 10GBase-T. Bruce Tolley of 10GbE silicon vendor Solarflare said this issue is being addressed and predicted that the likes of Cisco, Force10 and HP would be shipping 10GBase-T kit later this year.

“I think that the tier-one vendors will announce products and the expectation is that once we start shipping 10GBase-T systems and the switch and server adapter people start supporting this, then we’ll be looking at a hockey stick on the 10Gbit side,” he said.

Tolley said the issue of power consumption would shortly be resolved. “In 2005 we had a 20W part that gave 10Gbit/s over 50 metres. The part we sampled in 2006 used half the power at 10W and gave 10Gbit/s over 100 metres. The next product we announce will have the power output down by half again,” he said.