Hi, Here is the link to video about QEMU's recompilation engine: https://chemnitzer.linux-tage.de/2012/vortraege/1062
Short summary is that qemu can be used to emulate any CPU and can run binaries not targeted to native platform (e.g. any ELF executable complied for PowerPC can be executed on x86_64 using qemu-ppc). So to measure the overhead of the runtime translation, I used apache-bench (ab) tool to measure the transactions per second executed by the Tornado webserver (written in Python) which is being run in transtaed mode using qemu-x86_64. Tornado webserver throughput when executed natively was 1200-1400 transactions per seconds. But under qemu-x86_64, the throughput reported was 137-140 transactions per seconds. Why is this the case given that qemu chains the transalation blocks to avoid juming back and forth between translation block & translation loop? Is there any way to improve the performance of webserver or qemu? Thanks, Shashank Khasare
