Hmmm, so much of this looks like an attempt to solve the problems that were solved with fibre channel a couple decades back. Which I guess is standard NIH, with the advantage of not having to pay the FC consortium 95% HW margins. But still, you would think that some of those lessons could be learned before replacing it. AKA FC routes IP as one of its many protocols on top of the lower levels providing far more servic…
Maybe a better example would be Infiniband which is a simple and efficient protocol... but it's basically owned by Nvidia. For whatever reason Broadcom won't make Infiniband ASICs and Google doesn't want to be locked in to Nvidia so they have to use Ethernet.
Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
121–130 of 158 posts
Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#122Earlier quoted context omitted.
Which part of falcon, specifically, do you think duplicates off-the-shelf software that accomplishes the same things? Swift? Carousel? PLB?
There seems to be quite a lot of overlap between Falcon and what the Ultra Ethernet Consortium is ostensibly working on. As well as Amazon's Scalable Reliable Datagram (SRD) thing. All of them, in a way, are about addressing deficiencies in RoCEv2 for large scale latency sensitive networking that you see in HPC and DL training. But none of these are things you can buy today. Well, there's InfiniBand, but if you're we…
Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#123Oh, I was hoping this would be something built more directly over Ethernet, rather than on top of UDP/IP (if I'm understanding the layer diagram correctly). I've been working with Ethernet devices a lot lately, using the network as a communication bus, essentially. I find that there's a lot of complexity that we simply don't need: ARP, DHCP, DNS... So many points of failure. We know all the devices on our LAN and the…
Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#124Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#125Oh, I was hoping this would be something built more directly over Ethernet, rather than on top of UDP/IP (if I'm understanding the layer diagram correctly). I've been working with Ethernet devices a lot lately, using the network as a communication bus, essentially. I find that there's a lot of complexity that we simply don't need: ARP, DHCP, DNS... So many points of failure. We know all the devices on our LAN and the…
Ah yes, instead of going to google.com or 192.168.1.1 or adding a printer connected to my Wi-Fi, let me open my big yellow pages of globally unique MAC addresses…
Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#126Is it just me getting older / less smart, or did articles about products really start to sound like a jumbled mess or buzzwords lately? What is "Hardware transport", what is "the ecosystem"? And then there is dozens of random products and technologies that I've never heard of... This sounds more like a humble brag, than an article trying to inform people about technologies that might actually be useful to them.
This is how you get promoted at Google.
As we all know, interest at Google will now wane since you can't get promotions out of this any more.
Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#127Oh, I was hoping this would be something built more directly over Ethernet, rather than on top of UDP/IP (if I'm understanding the layer diagram correctly). I've been working with Ethernet devices a lot lately, using the network as a communication bus, essentially. I find that there's a lot of complexity that we simply don't need: ARP, DHCP, DNS... So many points of failure. We know all the devices on our LAN and the…
It's been done. There's XNS, and there's QNX networking over raw Ethernet. Both worked well. Look them up. You probably don't want to go that route, but it is technically possible on a LAN.
I believe there are other broadcast protocols that runs on Ethernet for the same reason.
Fibre Channel-over-Ethernet used to be a thing too, but I haven't seen it for a while. Perhaps latency wasn't as much of an issue as people thought and it lost to iSCSI.
I don't think it necessarily is bad idea to run protocols directly on the data link layer. The fewer parts the better. It's just that somewhere someone probably wants to route it, and the more general usage tends to win. So it's always going to be a niche market where latency is really important.
Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#128When you have enough scale you can claim a certain particular way of doing things are better than the others, which in most cases is just one way of doing things. This is what we see here.
Same thing goes for their single monolith repository philosophy.
Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#129Oh, I was hoping this would be something built more directly over Ethernet, rather than on top of UDP/IP (if I'm understanding the layer diagram correctly). I've been working with Ethernet devices a lot lately, using the network as a communication bus, essentially. I find that there's a lot of complexity that we simply don't need: ARP, DHCP, DNS... So many points of failure. We know all the devices on our LAN and the…
Stuff like InfiniBand, HPE Slingshot, Atos BXI, ... There is a consortium that's building a specification for those kinds of things: https://ultraethernet.org/
Re: Google opens Falcon, a reliable low-latency hardware transport, to the ecosystem
#130Earlier quoted context omitted.
This is exaggerated to the point that I consider it fiction. BTW Google doesn't substantially use gRPC within their datacenters. The industry has always been this way. Back in the day there were many many processor ISAs that are now consolidated. There was been many networking standards that consolidated (IPX/SPX anyone?) New things often diverge because of new requirements not out of spite. There is a push and pull…
> This is exaggerated to the point that I consider it fiction. BTW Google doesn't substantially use gRPC within their datacenters. For explicitness Google uses stubby which shares a lot of interface level commonality with gRPC but there's differences at a runtime level. Nobody is slinging json or soap around Google data centers.
Japanese Bullet Train technology