I find the "no toe, no way" attitude strange. I've seen a number of server applications that: a] move a lot of data over TCP.. let's say around 1 Gbps over a hundred concurrent flows. b] spend a significant amount of cycles in the kernel stack doing this. c] spend the rest of their cycles doing userspace crunching d] have latent unsatisfied demand.
Clearly this box needs more cycles.. If it can add a TOE and move some of b->c that is a pretty cheap and easy way of getting ahead and satisfying at least some of d. The issue is not that the box can't do 1Gbps when doing nothing else.. the issue is that it takes significant cycles to do 1Gbps. If I have to upgrade the general purpose processors 1] I may lose my existing capital investment 2] If I'm at a boundary I might have to add processors (turn a UP into an SMP or a 2-way into a 4-way) each of which add significant complication plus extra heat when compared to the TOEs.. not all scenarios are like this, and I agree TOEs are over-pitched.. but I think they certainly play a role in whole system design decisions that are bigger than just the kernel. While the scale of a,b,c, and d are going to change over time I don't really see the balance shifting any. -Patrick - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
