Archives


- Beowulf
- Beowulf Announce
- Scyld-users
- Beowulf on Debian

[Beowulf] about hyperthreading in HPC

Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.

Search

Mark Hahn hahn at physics.mcmaster.ca
Mon Aug 22 09:37:31 PDT 2005


> I have a cluster with 6 nodes Intel Xeon processors with Hyperthreading
> technology. My question is about if this technology has advantages on HPC
> clustering, or deppends of the application only?.

depeneds entirely on the application.  hyperthreading is nothing more than
hardware-assisted time-slicing.  namely, the CPU pretends that it's two 
CPUs, though in reality there is still a single pipeline.  one thread 
runs until it hits something like a cache miss or slow instruction, then 
the other thread runs for a while.  this can lead to moderately improved
throughput, but never to faster single-thread performance.  and often
the threads will conflict (demanding the same resource, such as competing 
fro cache).  when that happens, you can see _lower_ throughput...

> Then i found an article here:
> http://www.daemonology.net/hyperthreading-considered-harmful/ that talks about
> a security problem on servers with this technology. I wannt to know your
> opinions about this.

it's ridiculously mis-focused, and not all that novel.  mis-focused because
anyoen who can actually run malicious code on your cluster will probably 
find easier ways to own you.  not all that novel because there is a very long
history of this sort of timing-based information leaks; no one is interested
except perhaps governments.  and they have easier ways to go about it.

it would have been a lot more interesting during the cold war ;)




More information about the Beowulf mailing list