panhow commented on issue #9343: URL: https://github.com/apache/apisix/issues/9343#issuecomment-1519693397
> I can only provide some reference based on my experience. > > When I directly load test an HTTP server written in Go (only outputting 'hello world'), the QPS is 89246, while when forwarded by AIPSIX( the configuration of APISIX server is 8 cores and 8GB RAM. ), the QPS is 72765. The concurrency during load testing was 500. From my results, it seems that there is not a significant difference in the load testing results between using APISIX forwarding and directly load testing the target server. > > I think the following points can be considered: > > 1. The bandwidth situation of AIPSIX nodes, investigate whether there are any bottlenecks. > 2. Check the kernel parameters of APISIX's Linux node. > > Below are my system kernel parameters of APISIX node, you can take a look for reference( do not use directly for production due to differences in network environment. )： > > ``` > kernel.msgmnb = 655360 > kernel.msgmax = 65536 > kernel.msgmni = 16384 > kernel.shmmax = 68719476736 > kernel.shmall = 4294967296 > net.ipv4.ip_forward = 1 > net.ipv4.conf.all.rp_filter = 0 > net.ipv4.conf.default.rp_filter = 0 > net.ipv4.conf.all.accept_source_route = 0 > net.ipv4.conf.default.accept_source_route = 0 > net.ipv4.tcp_tw_recycle = 0 > net.ipv4.tcp_tw_reuse = 1 > net.ipv4.tcp_timestamps = 0 > net.ipv4.tcp_syncookies = 1 > net.ipv4.conf.lo.arp_ignore = 1 > net.ipv4.conf.lo.arp_announce = 2 > net.ipv4.conf.all.arp_ignore = 1 > net.ipv4.conf.all.arp_announce = 2 > net.ipv4.tcp_retrans_collapse = 1 > net.ipv4.conf.all.log_martians = 0 > net.ipv4.conf.macv-host.log_martians = 0 > net.ipv4.conf.default.log_martians = 0 > net.ipv4.conf.bond0.log_martians = 0 > net.ipv4.ip_nonlocal_bind = 0 > fs.inotify.max_queued_events = 16384000 > fs.inotify.max_user_instances = 1280000 > fs.inotify.max_user_watches = 8192000 > net.core.netdev_max_backlog = 200000 > net.bridge.bridge-nf-call-iptables = 1 > net.bridge.bridge-nf-call-ip6tables = 1 > vm.swappiness = 0 > vm.overcommit_memory = 1 > vm.panic_on_oom = 0 > fs.file-max = 52706963 > fs.nr_open = 52706963 > net.netfilter.nf_conntrack_tcp_timeout_established = 432000 > net.nf_conntrack_max = 10485760 > net.netfilter.nf_conntrack_max=10485760 > net.netfilter.nf_conntrack_buckets = 655360 > net.ipv4.neigh.default.gc_thresh1 = 163840 > net.ipv4.neigh.default.gc_thresh2 = 327680 > net.ipv4.neigh.default.gc_thresh3 = 500000 > net.ipv6.neigh.default.gc_thresh1 = 163840 > net.ipv6.neigh.default.gc_thresh2 = 327680 > net.ipv6.neigh.default.gc_thresh3 = 500000 > kernel.pid_max = 1966080 > kernel.threads-max = 2062606 > vm.max_map_count = 26214400 > net.core.somaxconn = 2621440 > net.ipv4.tcp_max_syn_backlog = 3276800 > net.ipv4.tcp_max_orphans = 2621440 > net.ipv6.conf.all.disable_ipv6 = 1 > net.core.rmem_default = 212992 > net.core.wmem_default = 212992 > net.core.rmem_max = 16777216 > net.core.wmem_max = 16777216 > net.ipv4.tcp_rmem = 4096 87380 16777216 > net.ipv4.tcp_wmem = 4096 65536 16777216 > net.ipv4.ip_local_port_range = 10000 65000 > net.ipv4.tcp_synack_retries = 5 > net.ipv4.tcp_syn_retries = 5 > net.ipv4.tcp_keepalive_time = 150 > net.ipv4.tcp_fin_timeout = 15 > net.ipv4.tcp_max_tw_buckets = 1440000 > net.ipv4.tcp_window_scaling = 1 > net.ipv4.tcp_sack = 1 > vm.min_free_kbytes = 262144 > vm.panic_on_oom = 0 > vm.vfs_cache_pressure = 200 > vm.swappiness = 30 > net.ipv4.route.max_size = 5242880 > net.ipv4.tcp_syn_retries = 6 > net.ipv4.tcp_retries1 = 3 > net.ipv4.tcp_retries2 = 6 > ``` > > Additionally, you can try increasing the number of worker_connections( vim config.yaml ): > > ``` > max_running_timers: 40960 # increase it if you see "lua_max_running_timers are not enough" error > > event: > worker_connections: 655350 > ``` summary above this: per core process 9000 request per second, in @githubxubin 's case, max cpu usage up to 45% of a 4 core machine, totally qps: 26000 request per second, it's about 26000/(4*0.45) = 14000 request per second, so this test make sense? maybe @githubxubin should figure out why apisix can't run cpu usage up to 100% -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
