On Tue, 2009-03-03 at 02:17 -0800, New User wrote: > Let me rephrase my first question below. What I would like to know is how > well can LVS code scale with multiple CPU. I do not know that balancing out > IRQ with the NIC works pretty well with Linux 2.6.18 kernel. Certain Broadcom > NICs seem to have problem with MSI turned on, but Intel NICs works fine. So, > *assume* the kernel can linearly scale up with SMP with NICs' interrupts, can > the LVS code do the same?
As it's a part of the kernel, the answer is "yes". However the more stuff you ask the kernel to do (ie. the less simple the packet path is through the stack), the lower the throughput will be. It this was a simple router you'll get throughput A pps. Adding LVS on top reduces A depending, again, on complexity - persistence maps, timeouts, the number of realservers. Adding (for example) netfilter/iptables rules reduces A further, and we all know that adding in conntrack features or modules will reduce it even further. One of the problems Linux has in general is that no two systems are alike, so benchmarking things is almost impossible in reality. I'd be very interested to see if you make your interfaces have affinity with a single CPU core whether your usage drops, because once they're locked they're avoiding softirq... Graeme _______________________________________________ Please read the documentation before posting - it's available at: http://www.linuxvirtualserver.org/ LinuxVirtualServer.org mailing list - [email protected] Send requests to [email protected] or go to http://lists.graemef.net/mailman/listinfo/lvs-users
