Why is my webcrawler throughput so low when parallel connections is over 10K?

Question

I have implemented a webcrawler with libcurl and libev. My intention was to make a high performance crawler that uses all available bandwidth. I have succeeded in making a c