ROBERT BLUM SILICON PHOTONICS
Intra Data Centre Network Topology
Core Network / Inter DataCenter
Super Spine/Core
Super Spine/Core
additional ICs- as well as increased complexity, but it does allow the reuse of existing fibre infrastructure already deployed for 100G PSM4. For duplex fibre, no uncooled option currently exists and it’s to be expected that MSAs will form to address the need for a low-cost and manufacturable duplex fibre solution.
multiple active alignment steps. And discrete designs also drive larger footprints, which limits their scalability to future form factors. SWITCHING TRENDS The TOR, leaf and spine switches being deployed today are typically 3.2Tb/s Ethernet switches in a 1RU chassis, though some 2RU systems are also available at 6.4Tb/s switch capacity. These switches have 25G SERDES and work well with the 100G QSFP28 transceivers. As switches move from 3.2T/6.4T to 12.8T, line rates will move to 50G SERDES using PAM4 modulation. At these data rates, new transceivers based on 50G electrical I/O are needed. To support 12.8T of switch capacity in a single RU, 400G transceivers in a form factor about the size of the QSFP are needed and two MSAs have formed to do just that: the QSFP-DD (DD stands for double density) and the OSFP (O stands for octal). Both MSAs support 8 lanes of electrical I/O at 50G PAM4 and can therefore support 400G optical interfaces. The challenge will be which optical interface will be suitable for 400G data centre connectivity. The IEEE has standardised a DR4 interface, which is similar to 100G PSM4 but using 100G PAM4 optical modulation instead of 25G NRZ on 4 parallel fibres. Using PAM4 modulation results in a significantly reduced link budget and higher power consumption due to
Spine
Spine
Leaf Leaf
ToR
ToR*
*Optical connections that can be hundreds of meters and are giving to single mode as data centres upgrade to 100G connectivity
BEYOND 400G As data centres evolve beyond 12.8T switches and 400G pluggables, integration of optics with networking silicon will eventually become inevitable. This is indeed one of the benefits of integrating lasers, modulators and optical multiplexers on a single chip: It enables the continued evolution towards higher bandwidth and smaller form factors. This integration of photonics and networking silicon will likely be done in a multi-chip package, leveraging the developments made in semiconductor packaging and taking advantage of using the best process nodes for each component.
Today’s operational focus remains on delivering 100G transceivers for data centres which already drive volumes in the millions of units per year. The next generation 400G pluggable form factors are defined and the industry is preparing to demonstrate 12.8T switches with optical interfaces in the near future. As we look beyond 12.8T, pluggable transceivers are likely running out of steam – they are too large and too power hungry – and we expect a paradigm shift in how optics will be deployed in the data centre. Robert Blum is Director of Strategic Marketing and Business Development at Intel’s Silicon Photonics Division.
Providing Complete Measurement Solutions for the Datacenter
100G PAM-4
OPTICAL NETWORKING SOLUTIONS
400G PAM-4
COHERENT OPTICAL
NETWORKING
SERVER
STORAGE
tek.com/Datacom
xxxxxxx xxxxxxx xxxxxxx xxxxxxx xxxxx
34
| ISSUE 10 | Q3 2017
www.opticalconnectionsnews.com
Made with FlippingBook - Online magazine maker