PostgreSQL 9.6.2 perfomance

I need help for server side optimization. I think it is a system file cache problem but I'm not sure, maybe I am wrong. I have 2 servers (S1, S2) with PostgreSQL 9.6.2 on both of them, the database contents are same. Profiling query: S1 Limit (cost=0.28..7.42 rows=1 width=32) (actual time=0.035..0.035 rows=1 loops=1) -> Nested Loop (cost=0.28..14.56 rows=2 width=32) (actual time=0.035..0.035 rows=1 loops=1) Join Filter: (djangocms_blog_blogcategory.id = djangocms_blog_post_categories.blogcategory_id) Rows Removed by Join Filter: 10 -> Index Only Scan using djangocms_blog_post_categories_post_id_blogcategory_id_key on djangocms_blog_post_categories (cost=0.28..12.28 rows=2 width=4) (actual time=0.017..0.017 rows=1 loops=1) Index Cond: (post_id = 20) Heap Fetches: 1 -> Materialize (cost=0.00..1.45 rows=30 width=32) (actual time=0.012..0.014 rows=11 loops=1) -> Seq Scan on djangocms_blog_blogcategory (cost=0.00..1.30 rows=30 width=32) (actual time=0.008..0.008 rows=11 loops=1) Planning time: 0.539 ms Execution time: 0.065 ms S2 Limit (cost=0.28..7.42 rows=1 width=32) (actual time=0.059..0.060 rows=1 loops=1) -> Nested Loop (cost=0.28..14.57 rows=2 width=32) (actual time=0.059..0.059 rows=1 loops=1) Join Filter: (djangocms_blog_blogcategory.id = djangocms_blog_post_categories.blogcategory_id) Rows Removed by Join Filter: 10 -> Index Only Scan using djangocms_blog_post_categories_post_id_blogcategory_id_key on djangocms_blog_post_categories (cost=0.28..12.29 rows=2 width=4) (actual time=0.032..0.032 rows=1 loops=1) Index Cond: (post_id = 20) Heap Fetches: 1 -> Materialize (cost=0.00..1.45 rows=30 width=32) (actual time=0.012..0.021 rows=11 loops=1) -> Seq Scan on djangocms_blog_blogcategory (cost=0.00..1.30 rows=30 width=32) (actual time=0.006..0.008 rows=11 loops=1) Planning time: 0.847 ms Execution time: 0.126 ms Why are the above execution times different approximately ~ 2 times slower on S2 server. System info: S1: usage hardware storage like /dev/sda OS: CoreOS 1185.3.0, Linux 4.7.3-coreos-r2 CPU: Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz RAM: 8 banks DIMM Synchronous 2133 MHz (0.5 ns), vendor: Micron, size: 16GiB Motherboard: ASUSTeK COMPUTER INC. Z10PA-U8 Series S2: usage software RAID 1 like /dev/md OS: Ubuntu 16.04, Linux 4.8.0-49-generic CPU: Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz RAM: 8 banks DIMM Synchronous 2400 MHz (0.4 ns), vendor: Micron, size: 31GiB Motherboard: ASUSTeK COMPUTER INC. Z10PA-U8 Series sysctl -a differences between S1 and S2 --- S1 2017-05-10 12:06:13.000000000 +0400 +++ S2 2017-05-10 12:06:17.000000000 +0400 @@ -2,0 +3,30 @@ +debug.kprobes-optimization = 1 @@ -3,0 +34,5 @@ +dev.mac_hid.mouse_button2_keycode = 97 +dev.mac_hid.mouse_button3_keycode = 100 +dev.mac_hid.mouse_button_emulation = 0 +dev.raid.speed_limit_max = 200000 +dev.raid.speed_limit_min = 1000 @@ -6 +41 @@ -fs.aio-nr = 16248 +fs.aio-nr = 0 @@ -8 +43 @@ -fs.dentry-state = 2479916 2429108 45 0 0 0 +fs.dentry-state = 801876 778676 45 0 0 0 @@ -10,5 +45,5 @@ -fs.epoll.max_user_watches = 27012874 -fs.file-max = 13160039 -fs.file-nr = 7936 0 13160039 -fs.inode-nr = 1219692 585566 -fs.inode-state = 1219692 585566 0 0 0 0 0 +fs.epoll.max_user_watches = 54067363 +fs.file-max = 131070 +fs.file-nr = 2944 0 131070 +fs.inode-nr = 743397 373 +fs.inode-state = 743397 373 0 0 0 0 0 @@ -19,0 +55 @@ +fs.mount-max = 100000 @@ -25,10 +60,0 @@ -fs.nfs.idmap_cache_timeout = 0 -fs.nfs.nfs_callback_tcpport = 0 -fs.nfs.nfs_congestion_kb = 262144 -fs.nfs.nfs_mountpoint_timeout = 500 -fs.nfs.nlm_grace_period = 0 -fs.nfs.nlm_tcpport = 0 -fs.nfs.nlm_timeout = 10 -fs.nfs.nlm_udpport = 0 -fs.nfs.nsm_local_state = 3 -fs.nfs.nsm_use_hostnames = 0 @@ -49 +75 @@ -fs.quota.syncs = 0 +fs.quota.syncs = 738 @@ -52,2 +77,0 @@ -fscache.object_max_active = 12 -fscache.operation_max_active = 6 @@ -62 +86 @@ -kernel.core_pattern = |/usr/lib/systemd/systemd-coredump %P %u %g %s %t %c %e +kernel.core_pattern = core @@ -64 +88 @@ -kernel.core_uses_pid = 1 +kernel.core_uses_pid = 0 @@ -71,2 +95,3 @@ -kernel.hardlockup_panic = 1 -kernel.hostname = db01 +kernel.hardlockup_panic = 0 +kernel.hostname = db02 +kernel.hotplug = @@ -77 +102 @@ -kernel.io_delay_type = 0 +kernel.io_delay_type = 1 @@ -81,0 +107 @@ +kernel.keys.persistent_keyring_expiry = 259200 @@ -86 +111,0 @@ -kernel.latencytop = 0 @@ -89,0 +115 @@ +kernel.moksbstate_disabled = 0 @@ -96 +122 @@ -kernel.ns_last_pid = 30114 +kernel.ns_last_pid = 5983 @@ -102 +128 @@ -kernel.osrelease = 4.7.3-coreos-r2 +kernel.osrelease = 4.8.0-49-generic @@ -109 +135 @@ -kernel.panic_on_stackoverflow = 0 +kernel.panic_on_rcu_stall = 0 @@ -114 +140 @@ -kernel.perf_event_max_sample_rate = 50000 +kernel.perf_event_max_sample_rate = 32000 @@ -117 +143 @@ -kernel.perf_event_paranoid = 2 +kernel.perf_event_paranoid = 3 @@ -121 +147 @@ -kernel.printk = 7 4 1 7 +kernel.printk = 4 4 1 7 @@ -122,0 +149 @@ +kernel.printk_devkmsg = ratelimit @@ -126 +153 @@ -kernel.pty.nr = 3 +kernel.pty.nr = 8 @@ -128,2 +155,2 @@ -kernel.random.entropy_avail = 825 +kernel.random.entropy_avail = 3413 @@ -133,2 +160,2 @@ -kernel.random.write_wakeup_threshold = 896 +kernel.random.write_wakeup_threshold = 1024 @@ -148 +175 @@ -kernel.sched_domain.cpu0.domain0.max_newidle_lb_cost = 7553 +kernel.sched_domain.cpu0.domain0.max_newidle_lb_cost = 12934 @@ -161 +188 @@ -kernel.sched_domain.cpu0.domain1.max_newidle_lb_cost = 8457 +kernel.sched_domain.cpu0.domain1.max_newidle_lb_cost = 14562 @@ -174 +201 @@ -kernel.sched_domain.cpu1.domain0.max_newidle_lb_cost = 5790 +kernel.sched_domain.cpu1.domain0.max_newidle_lb_cost = 8807 @@ -187 +214 @@ -kernel.sched_domain.cpu1.domain1.max_newidle_lb_cost = 7589 +kernel.sched_domain.cpu1.domain1.max_newidle_lb_cost = 11616 @@ -200 +227 @@ -kernel.sched_domain.cpu10.domain0.max_newidle_lb_cost = 5613 +kernel.sched_domain.cpu10.domain0.max_newidle_lb_cost = 16533 @@ -213 +240 @@ -kernel.sched_domain.cpu10.domain1.max_newidle_lb_cost = 5614 +kernel.sched_domain.cpu10.domain1.max_newidle_lb_cost = 15852 @@ -226 +253 @@ -kernel.sched_domain.cpu11.domain0.max_newidle_lb_cost = 5731 +kernel.sched_domain.cpu11.domain0.max_newidle_lb_cost = 9695 @@ -239 +266 @@ -kernel.sched_domain.cpu11.domain1.max_newidle_lb_cost = 6304 +kernel.sched_domain.cpu11.domain1.max_newidle_lb_cost = 12636 @@ -252 +279 @@ -kernel.sched_domain.cpu2.domain0.max_newidle_lb_cost = 4771 +kernel.sched_domain.cpu2.domain0.max_newidle_lb_cost = 12337 @@ -265 +292 @@ -kernel.sched_domain.cpu2.domain1.max_newidle_lb_cost = 6913 +kernel.sched_domain.cpu2.domain1.max_newidle_lb_cost = 11912 @@ -278 +305 @@ -kernel.sched_domain.cpu3.domain0.max_newidle_lb_cost = 5315 +kernel.sched_domain.cpu3.domain0.max_newidle_lb_cost = 8975 @@ -291 +318 @@ -kernel.sched_domain.cpu3.domain1.max_newidle_lb_cost = 12228 +kernel.sched_domain.cpu3.domain1.max_newidle_lb_cost = 12429 @@ -304 +331 @@ -kernel.sched_domain.cpu4.domain0.max_newidle_lb_cost = 7697 +kernel.sched_domain.cpu4.domain0.max_newidle_lb_cost = 11457 @@ -317 +344 @@ -kernel.sched_domain.cpu4.domain1.max_newidle_lb_cost = 7813 +kernel.sched_domain.cpu4.domain1.max_newidle_lb_cost = 12466 @@ -330 +357 @@ -kernel.sched_domain.cpu5.domain0.max_newidle_lb_cost = 5840 +kernel.sched_domain.cpu5.domain0.max_newidle_lb_cost = 6009 @@ -343 +370 @@ -kernel.sched_domain.cpu5.domain1.max_newidle_lb_cost = 6493 +kernel.sched_domain.cpu5.domain1.max_newidle_lb_cost = 12096 @@ -356 +383 @@ -kernel.sched_domain.cpu6.domain0.max_newidle_lb_cost = 6487 +kernel.sched_domain.cpu6.domain0.max_newidle_lb_cost = 9209 @@ -369 +396 @@ -kernel.sched_domain.cpu6.domain1.max_newidle_lb_cost = 11182 +kernel.sched_domain.cpu6.domain1.max_newidle_lb_cost = 13395 @@ -382 +409 @@ -kernel.sched_domain.cpu7.domain0.max_newidle_lb_cost = 6140 +kernel.sched_domain.cpu7.domain0.max_newidle_lb_cost = 9542 @@ -395 +422 @@ -kernel.sched_domain.cpu7.domain1.max_newidle_lb_cost = 5866 +kernel.sched_domain.cpu7.domain1.max_newidle_lb_cost = 15411 @@ -408 +435 @@ -kernel.sched_domain.cpu8.domain0.max_newidle_lb_cost = 8012 +kernel.sched_domain.cpu8.domain0.max_newidle_lb_cost = 7499 @@ -421 +448 @@ -kernel.sched_domain.cpu8.domain1.max_newidle_lb_cost = 7803 +kernel.sched_domain.cpu8.domain1.max_newidle_lb_cost = 14383 @@ -434 +461 @@ -kernel.sched_domain.cpu9.domain0.max_newidle_lb_cost = 5389 +kernel.sched_domain.cpu9.domain0.max_newidle_lb_cost = 9134 @@ -447 +474 @@ -kernel.sched_domain.cpu9.domain1.max_newidle_lb_cost = 6507 +kernel.sched_domain.cpu9.domain1.max_newidle_lb_cost = 12891 @@ -456 +483 @@ -kernel.sched_rr_timeslice_ms = 100 +kernel.sched_rr_timeslice_ms = 25 @@ -463,0 +491 @@ +kernel.secure_boot = 0 @@ -465,0 +494 @@ +kernel.sg-big-buff = 32768 @@ -473 +502 @@ -kernel.softlockup_panic = 1 +kernel.softlockup_panic = 0 @@ -476 +505 @@ -kernel.sysrq = 16 +kernel.sysrq = 176 @@ -478 +507 @@ -kernel.threads-max = 1030459 +kernel.threads-max = 2062506 @@ -482,0 +512,3 @@ +kernel.unprivileged_bpf_disabled = 0 +kernel.unprivileged_userns_apparmor_policy = 1 +kernel.unprivileged_userns_clone = 1 @@ -485 +517 @@ -kernel.version = #1 SMP Tue Nov 1 01:38:43 UTC 2016 +kernel.version = #52~16.04.1-Ubuntu SMP Thu Apr 20 10:55:59 UTC 2017 @@ -488,0 +521 @@ +kernel.yama.ptrace_scope = 1 @@ -499 +532 @@ -net.core.default_qdisc = fq_codel +net.core.default_qdisc = pfifo_fast @@ -514 +547 @@ -net.core.somaxconn = 65535 +net.core.somaxconn = 131070 @@ -2452 +2059 @@ -vm.lowmem_reserve_ratio = 256 256 32 +vm.lowmem_reserve_ratio = 256 256 32 1 @@ -2459 +2066 @@ -vm.mmap_min_addr = 4096 +vm.mmap_min_addr = 65536 HDParm tests S1 hdparm -Tt /dev/sda9 /dev/sda9: Timing cached reads: 23880 MB in 1.99 seconds = 11974.99 MB/sec Timing buffered disk reads: 1316 MB in 3.00 seconds = 438.57 MB/sec S2 hdparm -Tt /dev/md2 /dev/md2: Timing cached reads: 20508 MB in 2.00 seconds = 10262.05 MB/sec Timing buffered disk reads: 1532 MB in 3.00 seconds = 510.02 MB/sec The difference in hdparm tests show 16 % lower cached reads on S2, but I do not think this is a the main reason of the 2 times slower execution time on S2. All tests was repeated ~ 20 times on each server.