Data Centre Ethernet (DCE) is the great white hope of convergence, the single über-network across which all other protocols will flow, simplifying network component acquisition and operating costs - but it's not that simple. Ethernet is a fragile, unreliable base for such a role; it drops packets and message transfer time across …
Aren't we re-inventing the wheel ?
I seem to remember that the whole point of fibre channel over ethernet was that it would use standard ethernet networking that everyone knows and that can be bought cheaply and easily. Yet here we have standards groups working to add functionality to ethernet to make it comparible to fibre channel.
I know fibre channel isn't the cheapest, but aren't we risking loosing the whole point of using ethernet by bolting on all these extra bits ? You can bet you ain't gonna get all these extra bits for free across all ethernet manufacturers, so you're still going to have to buy special storage certified ethernet switches, which will cost money
There's been talk of throwing away the Internet and starting again. With all the extra bolt-ons to ethernet, shouldn't we do the same here too ?
Aren't these guys simply trying to re-invent RSVP and apply it to switches?
"Early 2010 is going to be DCE proof of concept time. If it works then the beginning of the end of the physical Fibre Channel era will have arrived. That will be momentous."
That's what they said about COBOL. There are still people out there using Token Ring ffs ! (granted - not in a datacenter (caveat: that I'm aware of)).
I can just imagine the support calls for this sort of development once it's up and running. There will always be clients using the latest and greatest, but there are also a lot of people out there who want to own their bit of cabling 100%, not pass it on to some datacenter back-plane, lossless or not, especially for Database SAN. Can you guarantee that no-one is 'peeking' at this back-plane traffic ? Not dissing it though, just don't think it's the all-singing all dancing fandabidoozie-wotsit portrayed in the article.
"Intercept" DCE products are already here
There are DCE capable products on the market today. Cisco's Nexus 5000 is the best known at this point, and some other vendors claim to support "lossless" Ethernet and FCoE, which would suggest they support pre-standard DCE features.
One thing the article fails to mention is NICs will have to support these DCE protocols as well. Emulex's and QLogic's FCoE Converged Network Adapters (CNAs) support DCE protocols, Intel's 82598EB "Oplin" NICs support DCE protocols, and some other NIC vendors claim either support DCE features today or will support DCE features in the future with firmware upgrades.
DCE = Cisco
Nice summary of the IEEE activity. Note that "Data Center Ethernet" is a Cisco term and is not used by others. The IEEE work groups all fall under "Data Center Bridging Task Group" http://www.ieee802.org/1/pages/dcbridges.html and most vendors are using Converged Enhanced Ethernet (CEE).
Predictable latency - wow!
I'll believe that when I see it.
It's a shame that it's too late though - I've got to somehow synchronise data along a ~1.5km string of fibre switches later this year.
It would be easy if I had private hardware, but I don't - I'm sharing it with everyone else in the site.
Predictable latency would be lovely.
Even better would be switches that don't take 30 seconds to notice that I've plugged in a new device and start feeding it the multicast packets it needs.
I find it quite disturbing that the dirt-cheap Linksys is often better than a top-line Cisco for widespread multicast... Ah well.
This has been a problem since Ethernet
Reminds me of back in the 80s. We Brits used networking responsibly and designed traffic controls, those pesky Utah people just blasted out regardless. When pointed out that they were not behaving approriately, the answer was "we must be doing it right as we have the major market share so you are wrong".
They no longer have the market share! Unfortunately the arrival of the web and rapid take up resurrected the all but dead TCP/IP to become the main network transport, hence the rise of Cisco.
As for the analysis - you can get packet loss and drop on a point to point, recovery depends on the transport protocol, I spent many hours communicating how networks work to 3Com people. I haven't been involved in such a way for quite a few years now, so the new people are obviously not quite there yet.
The problems are similar to the ones facing us with the advent of remote bridges, then switches, then remote switches. The worst offender on slow routes used to be Microsoft causing broadcast storms at the drop of a hat, which meant slow lines were flooded with meaningless packets whilst real, important data was dropped as buffers overflowed. Long discussions resolved the problem but yet again the personnel have probably changed. Cisco are fairly good at what they do but were not too hot at innovation so the fact that they are spearheading is probably not a good thing.
The answer to it all is in the transport protocol, if there is a limited buffer count that holds until acknowledged, problem solved as everyone important shuts up while the contention dies down. If the "lost" traffic is important then it should be using a better protocol.
Blimey. If only the Ethernet camp had listened to those ATM upstarts about ten years ago. Never mind - everyone gets to revisit all the issues over another three-year round of standards committee meetings. Enjoy.
But WHERE'S the IT angle????
stuff the data centre
"Who's going to turn base Ethernet into gold?" stuff the data centre, when are the likes of RTL going to mass produce generic 10gigbit ethernet cards and switchs/router,to then sell them for £15 and £35/£55 a throw, or even all inclusive packs with CAT6 and switch/router for the SOHO and masses of home lan users ASAP.
thats were the gold is right now, i want cheap 10gigabit etherrnet yesturday....
AND, dont forget to also finally include generic windows driver bonding out the box, as todays linux has had for ages now, then theres also incentive to buy and bond your old RTL 1gigbit cards in windows too...
- Xmas Round-up Ten top tech toys to interface with a techie’s Christmas stocking
- Google embiggens its fat vid pipe Chromecast with TEN new supported apps
- Microsoft: Don't listen to 4chan ... especially the bit about bricking Xbox Ones
- Shivering boffins nail Earth's coldest spot
- Exploits no more! Firefox 26 blocks all Java plugins by default