Achieving *really* low IO times

[Borgmeister]

Hey, i have been asked by my boss to get a machine which will produce the fastest IO times. Its gotta be PC achictecture. So, im wondering, AMD or Intel (dual proc) which mobo(chipset and/or make model) and price is no object. (no earth simluators tho plz). Needs to have Dual Channel memory support and will be running a *highly* cut down version of linux.

can you please post some links to comparative data if possible.

Thanks

Borgmeister

 

[Matthias99]

Define "fastest IO times" better. Do you need high throughput, low response time, or both? What amount of storage?

If the dataset is small enough, you may be able to run the whole thing from a RAMDisk (esp. with 64-bit Linux and a system with 16GB of RAM or more). Can't get much faster than that.

If you need a *lot* fast storage, and price is no object, a SAN/NAS solution is probably your best bet.

 

[Borgmeister]

Both, we are taking data through one gigabit ethernet port-coverting the raw into XML. Then pumping it out ove another NIC

 

[Peter]

On affordable x86, you get best I/O throughput on properly designed Opteron boards. That's because (a) they each have three parallel HyperTransport links to do I/O with, and (b) because CPU-RAM and CPU-IO traffic don't have to squeeze through the same front side bus. There is little benefit to IO-RAM busmaster traffic and IO-CPU cache snoop activity inside the chipset, but the IO wins a lot from the multiple HT links - if the board is designed to make use of them.