JOHN WILLIAMSON FUTURE DATACENTRE
rigorous, industry attention is turning towards the use of 400G technology. “Inside the datacentre, emerging 400 Gigabit Ethernet optics will help alleviate the power and real estate challenges facing hyperscale datacentre operations, by doubling or quadrupling the data rate between switches and routers,” says Xenos. She also describes how one of the main advantages of 400G is that the technology offers better system performance and tuneable capacity, which is great for hyperscale datacentres, as well as double channel capacity across applications, an arrangement that can extend the life of fibre investments. For his part, Campbell views the progress towards 400G as an architectural upgrade that is by way of a continuation of the recent move from 40G to 100G fabrics. “With 400G switches, optical transceivers and copper cables becoming available, large datacentres can again upgrade,” he says. There are, however, impediments to the widespread deployment of datacentre 400G in the immediate term: the present expense and availability of the kit. “The cost of deploying transceivers is considerably high; this coupled with the fact that switch platforms are not readily available, means that the widespread commercial implementation of 400G is not likely to happen this year,” judges Anthony Clarkson, technical director at compatible transceiver supplier ProLabs. “Once the switches and routers are widely available, the volume of products will help to drive the cost down – making the rollout of 400G a more justifiable option.” But ready access to 400G is thought unlikely to signal the end of datacentre bandwidth expansion history. “400G optics is just the next stage of evolution in the datacentre “pipe” size. This is by no means the end of the cycle,” states Dr Nagarajan. “People are already talking about 800 Gbps and 1.6 Tbps optics inside the datacentres.” bandwidth and latency ladders may be, by themselves they aren’t the only factors transforming today’s datacentre networking landscape . Datacentre architectures are changing. “Higher radix and highly interconnected datacentres are being designed,” points out Dr Nagarajan. “This allows for smaller data pipes and larger interconnected mesh architectures.” Improved connectivity and scalability are two key components that need to be addressed to keep up with the volume of requests likely to be facing hyperscale datacentres. “With improved speed to ensure instant connection, the ability to rapidly scale in order to meet demands when there are spikes in traffic will allow datacentres to offer flexibility DATACENTRE SPREAD As important as moves up the
QSFP-DD GEES UP 400G? Many experts believe that Quad Small Form Factor Pluggable Double Density (QSFP-DD) modules will play a pivotal role in the wider adoption of 400G in datacentre networking. ProLab’s Clarkson predicts that QSFP-DD will help to drive costs down as it will be the volume option for 400G. “With the ability to use QSFP28 and QSFP+ transceivers in QSFP-DD ports - in the same way you can use a 1G SFP in an SFP+ port - there will also be upgrade options available, addressing future unprecedented demands that datacentre operators may face,” he explains. “Other options like CFP8 will have their uses, for example in longer distance transport, but the lower power and familiar footprint of QSFP-DD will make it the form factor of choice for hyperscale datacentres both now and in the future.”
switching will allow for resource management for improved levels of utilisation, scalability, reliability and power efficiency. It will also improve the efficiency of high-performance computing towards exascale computing.” Dr Nagarajan is also on board with the disaggregation theme, stating that the trend of building a single, large, contiguous datacentre building is not sustainable. “Recent architecture changes involve a series of regionally interconnected datacentres, within the compute latency constraints, to emulate a “large”, logically connected datacentre,” he remarks. Longer term, Dr Nagarajan thinks there may be challenges concerning pluggable optics given that, as switch ASIC speeds increase, the faceplate density of a switch chassis becomes more of a premium, and pluggable sizes can only be reduced by so much. “The current generation of switch ASIC has a 12.8 Tbps capacity, and pluggable modules are still the favoured optical interconnect technology,” he acknowledges. “For the next generation 25.6 Tbps and 51.2 Tbps switches, people are talking about switch ASICs with co- packaged optics, to solve the faceplate density and power constraints.” In terms of greater component integration, Molex looks to be on the same page. “Silicon photonics will become more important as speeds continue to increase,” predicts Campbell. “Future switch designs will likely incorporate silicon photonics- based optical engines in a co-packaged format with the switch ASIC, effectively eliminating the need for pluggable optics.” Ciena is of the opinion that future datacentre networking solutions will need to have a wider focus in their functional remit. “Platforms must not only support open application programming interfaces, they will need to allow application and back-office integration efforts as well as the compatibility of vendor specific or agnostic third-party software components,” concludes Xenos. “Increased network health and status information will also be key.”
Helen Xenos, senior director, Portfolio Marketing, Ciena
to clients,” comments Xenos. “Without network scalability and re-configurability, a hyperscale datacentre simply can’t function.” Drew Martin, market manager at optical switch technology innovator HUBER+SUHNER Polatis, believes optical switching is set to play a key role in the future of datacentre networking as the shift towards automated networks and software defined networking gathers pace. “Optical switching applications will offer a host of benefits for datacentre operators who are moving towards network automation, including automated fibre management, self-service connectivity, zero touch provisioning and bandwidth on demand,” he ventures. CENTRE FORWARD What might be coming down the datacentre networking turnpike? As might be anticipated, projected scenarios differ somewhat depending on who you ask. Martin envisages more mileage for optical switching. “The future will see a disaggregation of datacentres and the rise of edge computing, in which optical switching applications will be paramount,” he reasons. “Optical
18
| ISSUE 14 | Q3 2018
www.opticalconnectionsnews.com
Made with FlippingBook flipbook maker