 |
 |
These are very nice results indeed! Ksoftirq handles bottom-halves of irqs, which in your case are mostly triggered by the network frames coming from your 10g interface and this is why it eats up so much CPU. Have you run your test with the .36 kernel with the normal "desktop" preemption mode? Can you confirm that your improvement doesn't come from your specific network device driver handling the network frames by polling the device instead of issuing interrupts? Cheers, Lukasz 2011/4/4 Charles G Waldman <[log in to unmask]>: > > > > > The major difference between the UltraLight kernel and the stock SL > > > kernel is that the SL kernel is configured with preemption enabled, > > > which is great for a user desktop, but not for a high-throughput > > > data server. > > > and is this the only difference? > > No. The UL kernel is built from newer sources (2.6.36), and there are a > few other config tweaks, but the PREEMPT_NONE is the change that is > really significant for server performace. > > > > We have found that on our test server (16-core server Dell 2950 with 6 > > > RAID6 shelves) with the SL kernel, under load, the "ksoftirqd" daemon > > > so the speed benchmark is done accessing all 6 raid6 arrays or just one > > raid? > > All 6. > > - Charles >