Home    Bloggers    Messages    Polls    Resources    Newsletter Signup   
Tw  |  Fb  |  In  |  Rss
John W. Verity

Homer Simpson & High-Speed Servers

John W. Verity
Page 1 / 2   >   >>
JohnVerity
JohnVerity
12/31/2012 7:41:16 PM
User Rank
Blogger
Torus in a teacup
Did you know, a teacup, with its looped handle, is a torus, too? It is a donot that has been transformed, morphed in a way that relies on the elasticity of its surface.

50%
50%
JohnVerity
JohnVerity
12/24/2012 4:30:29 PM
User Rank
Blogger
Re: Virtualized IO
AMD's fabric network is handled with switching chips mounted on every motherboard. Together, these form a mesh, or fabric, that move packets around and switch them to where they need to go. The more m'boards, the more netweork bandwidth is available. This network will eventually stretch out of the chassis and connect to other chassis, the company tells me. But it needs to be understood that this network is not going to provide the same core-to-core latency that a multi-core chip can provide, for instance. This torus network is connecting a whole bunch of self-contained computers together, with all the overhead that entails.

50%
50%
SaneIT
SaneIT
12/17/2012 7:50:54 AM
User Rank
Peta
Re: Virtualized IO
I guess the question is how is AMD's solution addressing this with switches?  I can see how interconnects in a supercomputer could be designed to take advantage of this theory but I'm wondering how AMD gets from a star to a torus.

50%
50%
Michelle
Michelle
12/15/2012 4:42:25 PM
User Rank
Kilo
Re: super duper donut
That's impressive savings on an extremely efficient set up.

50%
50%
JohnVerity
JohnVerity
12/15/2012 2:11:42 PM
User Rank
Blogger
Virtualized IO
I think this AMD approach makes perfect sense: more compute power in less space and using less power and costing less, too. The thing to keep in mind is that not every job can run in such a server - which is, in fact, really a box of many more or less independent servers. They are good at running the LAMP stack, for instance, not the fancier stacks that one finds in high-end servers. AMD has managed to squeeze down the size of its mother boards for this machine by virtualizing much of the I/O - turning hardware into software.

50%
50%
Henrisha
Henrisha
12/15/2012 3:16:37 AM
User Rank
Kilo
Re: super duper donut
Agreed. It might seem like too much but to those who are already contemplating on getting new servers, it might be a pretty good deal that's hard to resist.

50%
50%
Henrisha
Henrisha
12/15/2012 3:14:28 AM
User Rank
Kilo
Re: super duper donut
That will be the day! When buying servers will feel like you're just buying donuts, and companies can throw in a free 13th server for your 'bulk' purchase. An entertaining and informative read at the same time. Thanks, John!

50%
50%
JohnVerity
JohnVerity
12/14/2012 9:03:46 AM
User Rank
Blogger
Re: super duper donut
Interesting post from last spring at SemiAccurate, a site about the chip market. It points out that SeaMicro's interconnect is actually a stripped-down, hot-rodded version of the PCIe bus:

 

[AMD] looked at what was needed for a server and picked out exactly what isn't needed for their uses. Anything that was not mandatory, they removed, turned off, or didn't implement with fanatical devotion. The features that were needed, things like SATA, keyboards, and networks ports, were deemed too power hungry to implement, so they didn't. Those were turned off too.

Instead of lots and lots of controllers, NICs, drives, and cables, SeaMicro put one in, their interconnect fabric. With some clever coding the rest of these bits, where absolutely necessary, are virtualized across the interconnect. All 768 individual servers in a SeaMicro box don't have 768 drives, they share one SAN. There aren't 768 NICs either, there are a handful of 10GigE ports. All of the components are picked to save as much power as possible too. This consolidation pays dividends, and that is added to the power saved by removing the physical parts. On top of this, the cabling and transfer power saved is immense too. It is 1/10th of a watt here, 1/2 watt there, but in the end, it adds up to a big number, a big number times 768.

 

50%
50%
Michelle
Michelle
12/14/2012 8:46:53 AM
User Rank
Kilo
Re: super duper donut
We aren't looking to buy 12 or 13 servers anytime soon but if we were a free 13th server would be fantastic!

50%
50%
anthony.h
anthony.h
12/14/2012 6:24:40 AM
User Rank
Peta
Re: super duper donut
13th Server ??? I dont think I can even come closer for something like that. Anyway a good combination of characters.

50%
50%
Page 1 / 2   >   >>
More Blogs from John W. Verity
A Twitter engineer has explained much of the company's approach to performance and triggered a fascinating debate about infrastructure as well.
Cloud-based computing offers major advantages that are increasingly difficult to ignore.
The chip colossus is slated soon to unveil a new design for server racks to be used in large-scale datacenters.
Companies running their own compute clouds are opting for the platform-as-a-service model as a way to make IT as a whole more agile and effective.
Serious questions are being raised about how, and if, society can cope as automation eliminates many well-paying jobs, as seems inevitable.
flash poll
follow us on twitter
like us on facebook
Datacenter Acceleration
About Us     Contact Us     Help     Register     Twitter     Facebook     RSS