Thanks Amos,
I am going to add something about it in the LoadBalancing as a
"recommendation" that will be explained in it.
It takes a while to write it down and organize it but I do think it will
add couple good technical notes.
These will help a cache implementor to have clear view over couple
important things while designing it.
Eliezer
On 11/07/2013 11:51 PM, Amos Jeffries wrote:
On 2013-11-08 08:59, Eliezer Croitoru wrote:
Now these instances will probably will talk HTCP or ICP, HTCP preferred.
Now I am wondering what will "cost" more?
using one Frontend instance that will LoadBalance the traffic or plain
HTCP?
I expect that kernel level load balancing and HTCP between the instances
would be faster.
If you have a single instance as frontend it has to either be non-SMP
(bottleneck!) or using UDS/HTCP/ICP/something to coordinate between its
own internal SMP workers.
So either way for the higher capacity setups you have some process
communication going on. But with frontend instance you have all the
parsing and processing overheads of that instance on top. The more
efficient you make the inter-process communication the worse the
frontend design looks.
Amos
