I'm more interested in what they're doing with NSEC3, which is a key loose end in IETF's boondoggle DNSSEC system. NSEC3 is to domain names in a zone transfer what crypt(3) is to passwords in /etc/shadow.
"None less than Daniel J. Bernstein (currently visiting TU/e) took the time to write dedicated software for our cluster of Core2 Quad CPUs." Why am I not surprised that djb's team won a crypto speed contest? :)
The Coding and Cryptography Computer Cluster is a a ten-node cluster of conventional desktop PCs. Each node has an Intel Core 2 Quad Q6600 CPU with a clock rate of 2.40GHz and direct fully cached access to 8GB of RAM. Each computer has a 750GB Western Digital SATA hard disk. The nodes are connected via switched Gigabit Ethernet using Marvell PCI-E adapter cards.
Read their tweets... "With 1 machine down (fan) still getting 428435640 hashes per second thanks to Dan's cool fast code." It would seem that DJB wrote some code for their cluster that was on par with GPU speeds.
I'm more interested in what they're doing with NSEC3, which is a key loose end in IETF's boondoggle DNSSEC system. NSEC3 is to domain names in a zone transfer what crypt(3) is to passwords in /etc/shadow.
I should have entered the contest, considering I have access to a 512 node hadoop cluster. Kicking myself for not entering now...
"None less than Daniel J. Bernstein (currently visiting TU/e) took the time to write dedicated software for our cluster of Core2 Quad CPUs." Why am I not surprised that djb's team won a crypto speed contest? :)
I hope he'll post more info about his cluster. The 2nd place guy was running 5 nVidia GPUs, so it would be interesting to compare performance.
From the bottom of the page:
The Coding and Cryptography Computer Cluster is a a ten-node cluster of conventional desktop PCs. Each node has an Intel Core 2 Quad Q6600 CPU with a clock rate of 2.40GHz and direct fully cached access to 8GB of RAM. Each computer has a 750GB Western Digital SATA hard disk. The nodes are connected via switched Gigabit Ethernet using Marvell PCI-E adapter cards.
It also mentions that he had help from 2 folks who contributed GPU time. I can't imagine that fairly small cluster would come close without luck.
Read their tweets... "With 1 machine down (fan) still getting 428435640 hashes per second thanks to Dan's cool fast code." It would seem that DJB wrote some code for their cluster that was on par with GPU speeds.
...is not a cloud-based solution. It's a homemade cluster.