freeradius on 64 bits

[John Dennis]

On 06/21/2009 06:39 PM, Sajeewa Warnakulasuriya wrote:
> Alan,
>
> Is there any advantage going with a 64bits system?

Others may have differing opinions or insights into issues I'm not aware 
of, but in the case of a radius server I don't believe a 64-bit system 
buys you all that much. The larger address space and expanded integer 
range is an advantage in some scenarios such as databases, clusters, 
etc. But the radius protocol and internal server operation doesn't 
really benefit from the wider integer and pointer addressing to the best 
of my knowledge. Nor would the backend database, the database tables for 
radius just aren't that large. One can't forget there is a performance 
penality for moving from 32-bits to 64-bits. Each situation is unique 
and it's dangerous to draw general conclusions but a rule of thumb 
around here is that compiled code (as opposed to interpreted code [1]) 
will consume an extra 1/3 of memory on average (that's run time locals 
and heap usage, plus extra for the larger code size). So you're paying a 
price to move larger amounts of data around and if you're not benefiting 
from the larger data then all you're doing is dropping your performance.

[1] The reason for the distinction between compiled code and interpreted 
code is that many interpreters always allocate the maximum size data 
element whereas a compiler won't automatically use the maximum size data 
element, it's smarter.

-- 
John Dennis <jdennis at redhat.com>

Looking to carve out IT costs?
www.redhat.com/carveoutcosts/