Re: Performances issues with SSD volume ?

> From: Thomas SIMON <tsimon(at)neteven(dot)com>
>To: Glyn Astill <glynastill(at)yahoo(dot)co(dot)uk>
>Cc: "pgsql-admin(at)postgresql(dot)org" <pgsql-admin(at)postgresql(dot)org>
>Sent: Tuesday, 26 May 2015, 15:44
>Subject: Re: [ADMIN] Performances issues with SSD volume ?
>
>I can't do bonnie++ now, beacause I've already prepared server as a
>slave, and I have no sufficient disk space available (program says File
>size should be double RAM for good results output)
>
>My current sar output on productions (HDD) server is:
>14:25:01 DEV tps rd_sec/s wr_sec/s
>avgrq-sz avgqu-sz await svctm %util
>
>11:05:01 vg_data-lv_data 954.37 2417.35 18941.77 22.38
>31.10 32.46 0.34 32.70
>11:15:01 vg_data-lv_data 1155.79 8716.91 21995.77 26.57
>25.15 21.74 0.40 46.70
>11:25:01 vg_data-lv_data 1250.62 6478.67 23450.07 23.93
>39.77 31.78 0.41 51.34
>11:35:01 vg_data-lv_data 842.48 2051.11 17120.92 22.76
>15.63 18.53 0.29 24.04
>11:45:01 vg_data-lv_data 666.21 1403.32 14174.47 23.38
>10.11 15.12 0.24 15.79
>11:55:01 vg_data-lv_data 923.51 6763.36 15337.58 23.93
>13.07 14.14 0.35 32.63
>12:05:01 vg_data-lv_data 989.86 9148.71 16252.59 25.66
>19.42 19.56 0.45 44.21
>12:15:01 vg_data-lv_data 1369.24 8631.93 24737.60 24.37
>35.04 25.54 0.45 61.33
>12:25:01 vg_data-lv_data 1776.12 7070.01 39851.34 26.42
>74.81 42.05 0.44 77.29
>12:35:01 vg_data-lv_data 1529.15 6635.80 85865.14 60.49
>54.11 35.34 0.48 72.89
>12:45:01 vg_data-lv_data 1187.43 4528.74 40366.95 37.81
>36.07 30.36 0.39 45.81
>12:55:01 vg_data-lv_data 984.48 3520.06 21539.36 25.45
>17.91 18.17 0.31 30.20
>13:05:01 vg_data-lv_data 926.54 6304.44 16688.94 24.82
>17.36 18.69 0.41 38.05
>13:15:01 vg_data-lv_data 1232.46 7199.65 29852.49 30.06
>40.17 32.53 0.42 51.60
>13:25:01 vg_data-lv_data 1223.46 3945.05 27448.15 25.66
>31.07 25.31 0.35 42.65
>13:35:01 vg_data-lv_data 1126.91 2811.70 22067.19 22.08
>24.33 21.55 0.32 36.00
>13:45:01 vg_data-lv_data 833.33 1805.26 17274.43 22.90
>24.40 29.25 0.30 25.41
>13:55:02 vg_data-lv_data 1085.88 7616.75 19140.67 24.64
>17.48 16.06 0.39 42.15
>14:05:01 vg_data-lv_data 691.52 3852.50 13125.53 24.55
>7.75 11.15 0.30 20.74
>14:15:01 vg_data-lv_data 1288.88 5390.41 24171.07 22.94
>33.31 25.76 0.36 46.31
>14:25:01 vg_data-lv_data 1592.88 3637.77 29836.89 21.02
>76.45 47.94 0.40 63.28
>14:35:01 vg_data-lv_data 1652.78 9502.87 31587.68 24.86
>58.97 35.58 0.44 72.46
>14:45:01 vg_data-lv_data 1623.82 6249.52 34148.46 24.88
>53.47 32.83 0.40 65.19
>14:55:01 vg_data-lv_data 1330.44 6516.11 26828.59 25.06
>55.66 41.81 0.42 55.46
>Average: vg_data-lv_data 1176.55 5508.02 26324.37 27.06
>33.86 28.72 0.39 45.59

So the i/o is read heavy, it would be interesting to see why that might be and some insight into running queries would go a long way there.

>Quick answer on principal values on db
>size ~ 150 GB.
>Standard comportement is 55% update, 40% select & 10 % insert ; around
>170 TPS.
>

Again, if this is a dedicated postgres server I'd be hoping for there to be pretty much 0 reads from disk as your whole database should fit in os buffer cache. Only 170 TPS, are these large operations like bulk loads and big selects?

Can you post the ouput of "free -m" and "cat /proc/meminfo"?