Experience with high speed switches for cluster

While I do not have direct experience with Infiniband it seems to be the most widely used practice for connecting compute cluster nodes. It has very low latency.
 
Quote from ghobgood:

While I do not have direct experience with Infiniband it seems to be the most widely used practice for connecting compute cluster nodes. It has very low latency.
Thanks for the response.

Right, that is why I need it. I was wondering about the Voltaire product line vs other vendors of Infiniband switches, but any experience with Voltaire would be welcome.

There are others out there, e.g.,

http://www.cisco.com/en/US/prod/col...s6421/product_data_sheet0900aecd8029fdf7.html

http://www.networkworld.com/news/2004/0510topspin.html

http://h20000.www2.hp.com/bizsuppor...n&cc=us&prodTypeId=329290&prodSeriesId=443866


nitro
 
its cool if you dont wanna share, but why the hell do you need that much bandwidth? you must be running one bad-ass app..
 
Quote from auspiv:
its cool if you dont wanna share, but why the hell do you need that much bandwidth? you must be running one bad-ass app..

OP requests info on switches for CLUSTER.


If you don't know what a cluster is, you can always ask.
 
Quote from Tums:

OP requests info on switches for CLUSTER.


If you don't know what a cluster is, you can always ask.

i know what a cluster is, and it may be just me starting out trading, but i'd like to know what type of a system requires that much bandwidth. i can imagine massive clusters that just comb through data looking for arb, or maybe pairs, but definitely something that takes lots of computing power.

i'd just like to hear the basics of what nitro has planned. or the basics of why he needs that much computing strength.

i'm not retarded or anything...
 
Quote from auspiv:

its cool if you dont wanna share, but why the hell do you need that much bandwidth? you must be running one bad-ass app..
Bandwidth is not the issue. It is latency.

nitro
 
Quote from auspiv:

i know what a cluster is, and it may be just me starting out trading, but i'd like to know what type of a system requires that much bandwidth. i can imagine massive clusters that just comb through data looking for arb, or maybe pairs, but definitely something that takes lots of computing power.

i'd just like to hear the basics of what nitro has planned. or the basics of why he needs that much computing strength.

i'm not retarded or anything...
We need a cluster because we will be trading hundreds of stocks on software that I wrote.

nitro
 
Quote from canyonman00:

Hmmm...what flavor of OS are you planning for the supercomputer design? Perchance it be on a Dell platform? :)
Huh?

I am not sure I follow what you asked. Dells are not OS platforms, but hardware (although I guess you can order hardware with OS pre-installed). Since I am not sure, I will just answer both. We will probably be running HP blades, e.g.

http://h71016.www7.hp.com/dstore/ctoBases.asp?ProductLineId=431&FamilyId=2822

and we will be running both Linux (FreeBSD and RedHat) and Windows on the HP cluster. HP has a partnership with Voltaire,

http://www.voltaire.com/download/datasheets/HP-REUT-VOLT-INTC.pdf

Another possibility is this:

http://www.supermicro.com/newsroom/pressreleases/2008/press032008.cfm

nitro
 
Back
Top