190 likes | 300 Vues
This study explores the impact of interrupts, specifically from high volumes of ICMP ping requests, on system performance. By comparing merged inclusive and exclusive time views against user-only versions, we identify approximately 16 seconds of dilation in MPSP exclusive time and 25 seconds in total inclusive time. The findings suggest that the application profile alone does not provide a complete picture. Key culprits of delays such as `schedule()`, `do_IRQ()`, and `icmp_reply()` are revealed, demonstrating the complexity of OS interactions in response to network traffic.
E N D
1. Observing Program-OS InteractionExample 1: "Interrupts” (Interrupts induced by large number of icmp pings) NAS Parallel Benchmark SP Class W run with and without interrupts User-Only Inclusive Time View User-Only Exclusive Time View Approx. 16 secs dilation in MPSP() Exclusive Time. Why? Approx. 25 secs dilation in Total Inclusive Time. Why? Application Profile does not tell the whole story!
1. Observing Program-OS InteractionExample 1: "Interrupts” (Interrupts induced by large number of icmp pings) Merged Inclusive Time View Merged Exclusive Time View MPSP excl. time difference only 4 secs. Excl-time view clearly identifies the culprits. 1. schedule() 2. do_IRQ() 3. icmp_reply() 4. do_softirq() Pings cause interrupts (do_IRQ). Which in turn handled after interrupt by soft-interrupts (do_softirq). Actual routine is icmp_reply/rcv. Large number of softirqs causes ksoftirqd to be scheduled-in, causing SP to be scheduled-out. Kernel-Space Time Taken by: 1.do_softirq() 2. schedule() 3. do_IRQ() 4. sys_poll() 5. icmp_rcv() 6. icmp_reply()