On 16/04/2016 6:41 a.m., Mohammad Sadegh Nasiri wrote: > Thanks Amos for your reply. > > > This reaches 800 Mbps with Squid still spending a measurable chunk of its >> time (~30%) waiting for something to do. > > How do you discovered to this numbers (800Mbps and ~30%)? I explained the 800 Mbps in my last post. "adding these together" right under the set of values which were added together. ~30% CPU is 100% CPU - 67% CPU. The difference of CPU rates in your traces. > > What is median_select_fds means? It is the average (median) number of sockets (FD) which the kernel tells Squid need some processing done (using CPU) when Squid performs an I/O select(2) to find that out. > > HW info: > Server model: ProLiant DL380 Gen8] > CPU: Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz > RAM: 128GB > Disks: Samsung SSD [1TB+512GB+512GB] > > I'm suspicious to Squid configuration or I/O (DMA, RAID) If RAID is having any effect it would be making your CPU baseline higher than it would be without RAID. And during a overload peak slowing down the recovery. It is unlikely to be the direct cause of the peak event. The bottleneck for your system appears to be concurrent transactions and CPU. The amount of work the CPU used by Squid can do each I/O cycle is less than the work the active FD are needing done. If you think it is configuration related, then show the config and I/we can point at things that are making that CPU work exist. Amos _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
