Continuous profiling: where have all the cycles gone?
ACM Transactions on Computer Systems (TOCS)
Diagnosing performance overheads in the xen virtual machine environment
Proceedings of the 1st ACM/USENIX international conference on Virtual execution environments
TotalProf: a fast and accurate retargetable source code profiler
CODES+ISSS '09 Proceedings of the 7th IEEE/ACM international conference on Hardware/software codesign and system synthesis
| Hi-index | 0.00 |
These days, the L4 microkernel is expanding its domain towards embedded systems since it is showing a comparable performance with traditional monolithic kernels. The L4 microkernel shows a greatly different execution behavior of user applications from that in a traditional monolithic environment because most operating-system services are run as user-level applications. Therefore, we need a profiling framework to obtain a better understanding of performance bottlenecks for software optimization. However, current L4 profiling tools provide only higher-level information, such as the number of function calls, IPCs, and context switches. In this paper, we present a software profiling framework which gathers system-wide statistical information in the L4 microkernel environment. In order to support profiling lower-level information such as clock cycles, cache misses, and TLB misses, our profiling framework uses the hardware performance counters of the PMU (Performance Monitoring Unit) which most CPUs support. In this paper, we show that our profiling framework incurs less than 3% overhead below 15000 interrupts per second compared to the existing Linux profiling tool. Moreover, as a case study, we show the main cause of performance loss in L4Linux applications compared with Linux applications.