What's In A Number? Er, Lots Of Packets Per Second...
I remember when first testing Gigabit Ethernet (Packet Engines and Foundry Networks for the record) and thinking: “how do we harness this much bandwidth – maybe take a TDM approach and slice it? After all, who needs a gig of bandwidth direct to the desktop?”
Well, that was still a relevant point as only servers were really being fitted with Gigabit Ethernet NICs in ’98/’99 and even then they could rarely handle that level of traffic. Especially the OS/2 based ones… However, as a backbone network tech it was top notch. In a related metric, testing switches and appliances such as Load-Balancers and getting throughput in packets per second terms in excess of 100,000 pps was mad, crazy – the future! So, Mellanox (newly in the hands of nvidia – trust me, the latter do more that make graphics cards!) has just released the latest of its Ethernet Cloud Fabric products with lots of “Gigabit Ethernet” in it – i.e. 100/200/400 GbE switches. I’ll match that and raise you 400…
And forget six figure pps throughput figures. How about 8.3bn pps instead? You can’t test that with 30 NT servers -)
If you’re wondering who needs that level of performance in the DC, think no further than the millions of customers being serviced by AWS, Azure and all the other CSPs – that’s a lot of traffic being handled by the cloudy types on your behalf. And then there’s research, gas and oil/exploration, FinTec, the M&E guys and, not least – given the nvidia ownership – extreme virtual reality. Bring on Terabit Ethernet…
Meantime, I think we’ll need a lot of 5G deployment at the other end of the data chain! But that’s another discussion for another day. And whatever did happen to WiMAX?