> I am not sure whether this is the right place for my > question. But I think it should be a general one. > > I am running netperf tcp_stream against Nevada build 78 on my x86 > system. This system is configured with a Intel Core2 processor and 2G > sized memory. I installed a dual ports Pro1000 NIC card and test the > tcp_stream performance through these two NIC ports. The first time I > run netperf on the native OS. The utilization of both ports are 100% > and the bandwidth of each is about 120MB/s. Then, I build and Cap-Eye > Installe onnv kernel from the b87 onnv source code. To my surprise, > the utilization of each ports reduces to 85% and the bandwidth is > 100MB/s in average (not balanced, say, one is 105 and another is 95). > > I thought the right way is to BFU the system, not just Cap-Eye install > the kernel. So I nightly and BFU the system. But the result is just > the same as the previous Cap-Eye Installed system. So what is the > difference between the native and BFUed Nevada? How can get th same > performance on the BFUed system?
When you compiled new kernel bits, did you compile them as "debug" or as "release" bits? I suspect you compiled debug bits; the debug kernel modules use quite a few extra asserts in the code, and they have things like kmem heap checking enabled. Most likely the heap checking is responsible for the slow down you're observing. This message posted from opensolaris.org _______________________________________________ opensolaris-discuss mailing list [email protected]
