3 ms·
To put this in perspective the largest Hadoop cluster Yahoo (or anyone) has spun up is 4000 2.5ghz dual-processor quad-core Xeons. According to http://www.clust
by sh1mmer 18y ago
To put this in perspective the largest Hadoop cluster Yahoo (or anyone) has spun up is 4000 2.5ghz dual-processor quad-core Xeons. According to http://www.clustervision.com/cluster_hardware.html http://www.clustervision.com/cluster_hardware.html a Xeon runs about 4 flops per cycle. That means the entire cluster was running at about 20 megaflops.
The smallest computer on that map is 27 teraflops. More than a factor of 1000 on our cluster.
That said, raw power isn't everything. I know we were all impressed with Google's recent TB processing record, which prove the old adage it's not how big it is, it's what you do with it.
- randomwalker 18y agoDamn. I think you managed to get every single number wrong :-) First of all, FLOPS is not the same as FLOPs / cycle. Note the lowercase 's'. The former is floating point operations per second. The latter is per cycle. To convert the latter number into FLOPS, you have to multiply by the clock speed. That gives you 10^10 FLOPS, or 10 gigaflops. Now that's per-core. You said quad-core, so it's 40 gigaflops per processor. 4000 processors, so it's 160 teraflops. Much higher than the slowest on that list, but much lower than the fastest. The other thing you got wrong is that if it were a mere 20 megaflops, it would have been a factor of a million slower :-) Tera is 10^12. Anyway, I think this whole thing is apples-to-oranges because actual supercomputers have a much more tightly inter-connected grid (I believe) than the sort of loose cluster that Hadoop typically runs on, and further, they have a different programming model (Fortran with MPI and things like that.)