[CC the xrootd list and some others who might be interested..]

  Hi Andy,

On Mon, Feb 28, 2005 at 10:19:03PM -0800, Andrew Hanushevsky wrote:
> We've just completed our performance tests with memory mapped I/O. The
> results were excellent. Server overhead decreased by 50% and client
> latency was decreased by at least 20%. 100% network utilization can
> now be easily reached. This is very good news for systems like proof.
> While memory mapped I/O produces excellent results it can only be used
> in specific circumstances where you have a lot of memory and a 64-bit
> address space. So, it is not on be default. Configuration is trivial
> and you have a lot of options, including being able to specify precisely
> which files should be memory mapped independently of the configuration
> file.

  Very nice: a network-attached distributed memory cache...


Peter Elmer     E-mail: [log in to unmask]      Phone: +41 (22) 767-4644
Address: CERN Division PPE, Bat. 32 2C-14, CH-1211 Geneva 23, Switzerland