Gets benchmarks running again and does a rough pass for 0.7.1.

8 years ago · 13d9476a43
5 changed files with 8683 additions and 69 deletions
--- a/bench/Makefile
+++ b/bench/Makefile
@ -1,4 +1,4 @@
-REQ=20480
+REQ=262144
 CLIENTS=64
 ADDR=http://127.0.0.1:8500/v1/kv/bench
 DATA="74a31e96-1d0f-4fa7-aa14-7212a326986e"
--- a/bench/README.md
+++ b/bench/README.md
@ -1,13 +1,14 @@
 Consul Benchmark
 ================

-This repo contains the Packer automation necessary for the Consul benchmarks.
+This repo contains the automation necessary for the Consul benchmarks.

 There is a single main Packer file `bench.json`. To use it, the variables
-for `do_client_id` and `do_api_key` must be provided. There correspond to
-your DigitalOcean client id and API key.
+for `do_client_id` and `do_api_key` must be provided. These correspond to
+your DigitalOcean client ID and API key.

 When Packer runs, it will generate 3 images:
+
 * bench-bootstrap - Consul server in bootstrap mode
 * bench-server - Consul server
 * bench-worker - Worker node
--- a/bench/bench.json
+++ b/bench/bench.json
@ -1,36 +1,31 @@
 {
  "variables": {
-        "do_client_id": "",
-        "do_api_key": ""
+    "do_size": "16gb",
+    "do_image": "ubuntu-14-04-x64",
+    "do_region": "nyc3"
  },
  "builders": [
    {
      "type": "digitalocean",
-        "api_key": "{{ user `do_api_key` }}",
-        "client_id": "{{ user `do_client_id` }}",
-        "region_id": "1",
-        "size_id": "66",
-        "image_id": "3101045",
+      "region": "{{ user `do_region` }}",
+      "size": "{{ user `do_size` }}",
+      "image": "{{ user `do_image` }}",
      "snapshot_name": "bench-bootstrap-{{ isotime }}",
      "name": "bootstrap"
    },
    {
      "type": "digitalocean",
-        "api_key": "{{ user `do_api_key` }}",
-        "client_id": "{{ user `do_client_id` }}",
-        "region_id": "1",
-        "size_id": "66",
-        "image_id": "3101045",
+      "region": "{{ user `do_region` }}",
+      "size": "{{ user `do_size` }}",
+      "image": "{{ user `do_image` }}",
      "snapshot_name": "bench-server-{{ isotime }}",
      "name": "server"
    },
    {
      "type": "digitalocean",
-        "api_key": "{{ user `do_api_key` }}",
-        "client_id": "{{ user `do_client_id` }}",
-        "region_id": "1",
-        "size_id": "66",
-        "image_id": "3101045",
+      "region": "{{ user `do_region` }}",
+      "size": "{{ user `do_size` }}",
+      "image": "{{ user `do_image` }}",
      "snapshot_name": "bench-worker-{{ isotime }}",
      "name": "worker"
    }
@ -46,9 +41,9 @@
      "inline": [
        "mkdir /etc/consul.d",
        "apt-get update",
-            "apt-get install unzip make",
-            "wget https://releases.hashicorp.com/consul/0.5.2/consul_0.5.2_linux_amd64.zip",
-            "unzip 0.5.2_linux_amd64.zip",
+        "apt-get install -y unzip make",
+        "wget https://releases.hashicorp.com/consul/0.7.1/consul_0.7.1_linux_amd64.zip",
+        "unzip consul_*_linux_amd64.zip",
        "mv consul /usr/local/bin/consul",
        "chmod +x /usr/local/bin/consul"
      ]
--- a/bench/results-0.7.1.md
+++ b/bench/results-0.7.1.md
@ -0,0 +1,223 @@
+# Consul Benchmark Results
+
+As part of a benchmark, we started a 4 node DigitalOcean cluster to benchmark.
+There are 3 servers, meaning writes must commit to at least 2 servers.
+The cluster uses the 16GB DigitalOcean droplet which has the following specs:
+
+ * 8 CPU Cores, 2Ghz
+ * 16GB RAM
+ * 160GB SSD disk
+ * 1Gbps NIC
+
+# Output
+
+Below is the output for a test run on a benchmark cluster. We ran the benchmark
+several times to warm up the nodes, and this is just a single representative sample.
+
+Note, that a single worker was running the benchmark. This means the "stale" test
+is not representative of total throughput, as the client was only routing to a
+single server.
+
+We also did an initial run where we got lots of noise in the results, so we
+increased the number of requests to try to get a better sample.
+
+```
+===== PUT test =====
+GOMAXPROCS=4 boom -m PUT -d "74a31e96-1d0f-4fa7-aa14-7212a326986e" -n 262144 -c 64 http://127.0.0.1:8500/v1/kv/bench
+262144 / 262144 Booooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo! 100.00 %
+
+Summary:
+  Total:	69.3512 secs.
+  Slowest:	0.0966 secs.
+  Fastest:	0.0026 secs.
+  Average:	0.0169 secs.
+  Requests/sec:	3779.9491
+  Total Data Received:	1048576 bytes.
+  Response Size per Request:	4 bytes.
+
+Status code distribution:
+  [200]	262144 responses
+
+Response time histogram:
+  0.003 [1]	|
+  0.012 [66586]	|∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
+  0.021 [146064]	|∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
+  0.031 [34189]	|∎∎∎∎∎∎∎∎∎
+  0.040 [9178]	|∎∎
+  0.050 [3682]	|∎
+  0.059 [1773]	|
+  0.068 [464]	|
+  0.078 [124]	|
+  0.087 [63]	|
+  0.097 [20]	|
+
+Latency distribution:
+  10% in 0.0095 secs.
+  25% in 0.0119 secs.
+  50% in 0.0151 secs.
+  75% in 0.0195 secs.
+  90% in 0.0260 secs.
+  95% in 0.0323 secs.
+  99% in 0.0489 secs.
+
+===== GET default test =====
+GOMAXPROCS=4 boom -n 262144 -c 64 http://127.0.0.1:8500/v1/kv/bench
+262144 / 262144 Booooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo! 100.00 %
+
+Summary:
+  Total:	34.8371 secs.
+  Slowest:	0.9568 secs.
+  Fastest:	0.0014 secs.
+  Average:	0.0085 secs.
+  Requests/sec:	7524.8570
+  Total Data Received:	36175872 bytes.
+  Response Size per Request:	138 bytes.
+
+Status code distribution:
+  [200]	262144 responses
+
+Response time histogram:
+  0.001 [1]	|
+  0.097 [261977]	|∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
+  0.192 [38]	|
+  0.288 [64]	|
+  0.384 [0]	|
+  0.479 [0]	|
+  0.575 [0]	|
+  0.670 [0]	|
+  0.766 [0]	|
+  0.861 [38]	|
+  0.957 [26]	|
+
+Latency distribution:
+  10% in 0.0044 secs.
+  25% in 0.0055 secs.
+  50% in 0.0072 secs.
+  75% in 0.0098 secs.
+  90% in 0.0130 secs.
+  95% in 0.0157 secs.
+  99% in 0.0228 secs.
+
+===== GET stale test =====
+GOMAXPROCS=4 boom -n 262144 -c 64 http://127.0.0.1:8500/v1/kv/bench?stale
+262144 / 262144 Booooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo! 100.00 %
+
+Summary:
+  Total:	26.8200 secs.
+  Slowest:	0.0838 secs.
+  Fastest:	0.0005 secs.
+  Average:	0.0065 secs.
+  Requests/sec:	9774.1922
+  Total Data Received:	36175872 bytes.
+  Response Size per Request:	138 bytes.
+
+Status code distribution:
+  [200]	262144 responses
+
+Response time histogram:
+  0.001 [1]	|
+  0.009 [214210]	|∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
+  0.017 [42999]	|∎∎∎∎∎∎∎∎
+  0.026 [3709]	|
+  0.034 [589]	|
+  0.042 [313]	|
+  0.050 [166]	|
+  0.059 [102]	|
+  0.067 [42]	|
+  0.075 [11]	|
+  0.084 [2]	|
+
+Latency distribution:
+  10% in 0.0031 secs.
+  25% in 0.0041 secs.
+  50% in 0.0056 secs.
+  75% in 0.0079 secs.
+  90% in 0.0109 secs.
+  95% in 0.0134 secs.
+  99% in 0.0203 secs.
+
+===== GET consistent test =====
+GOMAXPROCS=4 boom -n 262144 -c 64 http://127.0.0.1:8500/v1/kv/bench?consistent
+262144 / 262144 Booooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo! 100.00 %
+
+Summary:
+  Total:	35.6962 secs.
+  Slowest:	0.0826 secs.
+  Fastest:	0.0016 secs.
+  Average:	0.0087 secs.
+  Requests/sec:	7343.7475
+  Total Data Received:	36175872 bytes.
+  Response Size per Request:	138 bytes.
+
+Status code distribution:
+  [200]	262144 responses
+
+Response time histogram:
+  0.002 [1]	|
+  0.010 [183123]	|∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
+  0.018 [70460]	|∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
+  0.026 [6955]	|∎
+  0.034 [657]	|
+  0.042 [391]	|
+  0.050 [229]	|
+  0.058 [120]	|
+  0.066 [121]	|
+  0.074 [68]	|
+  0.083 [19]	|
+
+Latency distribution:
+  10% in 0.0047 secs.
+  25% in 0.0059 secs.
+  50% in 0.0077 secs.
+  75% in 0.0104 secs.
+  90% in 0.0137 secs.
+  95% in 0.0162 secs.
+  99% in 0.0227 secs.
+```
+
+# Profile
+
+In order to probe performance a bit, we ran the get-stale benchmark on the
+leader itself and collected pprof data. Here's the output of the benchmark:
+
+```
+===== GET stale test =====
+GOMAXPROCS=4 boom -n 262144 -c 64 http://127.0.0.1:8500/v1/kv/bench?stale
+262144 / 262144 Booooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo! 100.00 %
+
+Summary:
+  Total:        16.3139 secs.
+  Slowest:      0.0815 secs.
+  Fastest:      0.0001 secs.
+  Average:      0.0040 secs.
+  Requests/sec: 16068.7946
+  Total Data Received:  36175872 bytes.
+  Response Size per Request:    138 bytes.
+
+Status code distribution:
+  [200] 262144 responses
+
+Response time histogram:
+  0.000 [1]     |
+  0.008 [240221]        |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
+  0.016 [18761] |∎∎∎
+  0.025 [1937]  |
+  0.033 [496]   |
+  0.041 [293]   |
+  0.049 [131]   |
+  0.057 [162]   |
+  0.065 [127]   |
+  0.073 [10]    |
+  0.081 [5]     |
+
+Latency distribution:
+  10% in 0.0013 secs.
+  25% in 0.0019 secs.
+  50% in 0.0030 secs.
+  75% in 0.0046 secs.
+  90% in 0.0074 secs.
+  95% in 0.0109 secs.
+  99% in 0.0174 secs.
+```
+
+And here's the [resulting flame graph](results-0.7.1.svg).
--- a/bench/results-0.7.1.svg
+++ b/bench/results-0.7.1.svg