Skip to content

Latest commit

 

History

History
97 lines (82 loc) · 3.65 KB

BENCHMARK.md

File metadata and controls

97 lines (82 loc) · 3.65 KB

Benchmarks

Sanity's benchmarks were designed to compare the performance of Sanity to the performance of text/html. It is difficult fairly compare Sanity to text/html because their performance penalties are asymteric.

text/html uses reflection when interacting with the template's model. Reflection in Go is relatively slow, so the cost of most templates is driven by the number {{ }} blocks in the template. If a template contains no dynamic blocks, it compiles down to a single io.Writer.Write call and will outperform Sanity.

The internal design of Sanity was driven by a desire to minimize the number of allocations required to build a node tree, but even so every Sanity tag requires exactly one allocation and benchmarking Sanity is effectively benchmarking the Go allocator and garbage collector. So the cost of rendering a Sanity view directly tracks the number of tags rendered. But one allocation is much faster than one {{ }} block in a template, so most Sanity views will out perform an equivalent text/html template.

Micro Benchmarks

internal/benchmark contains micro benchmarks. Rendering a list with 1k elements requires 145us using Sanity and 803us using text/html.

goos: linux
goarch: amd64
pkg: github.com/jeffswenson/sanity/internal/benchmark
cpu: 11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz
BenchmarkSanityList1000-8           8078            145831 ns/op
PASS
ok      github.com/jeffswenson/sanity/internal/benchmark 1.201s

goos: linux
goarch: amd64
pkg: github.com/jeffswenson/sanity/internal/benchmark
cpu: 11th Gen Intel(R) Core(TM) i7-1185G7 @ 3.00GHz
BenchmarkGoTemplate1000-8           1501            803934 ns/op
PASS
ok      github.com/jeffswenson/sanity/internal/benchmark 1.295s

HTTP Benchmarks

internal/benchmarkhttp contains benchmarks that use the stdlib http server to render HTML content. The HTML content is similar to a popular orange tech news site. The benchmark contains tests with and without gzip.

The tests were run with GOMAXPROCS=1 and GOMEMLIMIT=100Mib. GOMAXPROCS was set to measure the single threaded throughput. The memory limit was set to to force frequent garbage collection. The tests saturated the server's capacity, so the latency numbers are meaningless.

A key takeway from the test is gziping the HTML costs far more cpu cyles than producing the HTML, so rendering the view is unlikely to be a bottleneck in most applications.

Bombarding http://localhost:8080/sanity for 10s using 125 connection(s)
Statistics        Avg      Stdev        Max
  Reqs/sec      8693.36     851.13    9551.49
  Latency       14.42ms    31.99ms   381.55ms
  HTTP codes:
    1xx - 0, 2xx - 86619, 3xx - 0, 4xx - 0, 5xx - 0
    others - 0
  Throughput:    92.30MB/s

Bombarding http://localhost:8080/sanity-gzip for 10s using 125 connection(s)
Statistics        Avg      Stdev        Max
  Reqs/sec      2698.47     462.90   11584.86
  Latency       46.44ms   187.45ms      2.05s
  HTTP codes:
    1xx - 0, 2xx - 26943, 3xx - 0, 4xx - 0, 5xx - 0
    others - 0
  Throughput:     2.40MB/s

Bombarding http://localhost:8080/template for 10s using 125 connection(s)
Statistics        Avg      Stdev        Max
  Reqs/sec      2230.99     353.45    8751.29
  Latency       56.25ms    23.46ms   129.58ms
  HTTP codes:
    1xx - 0, 2xx - 22228, 3xx - 0, 4xx - 0, 5xx - 0
    others - 0
  Throughput:    27.81MB/s

Bombarding http://localhost:8080/template-gzip for 10s using 125 connection(s)
Statistics        Avg      Stdev        Max
  Reqs/sec      1203.30      90.52    1350.24
  Latency      103.27ms   191.40ms      1.32s
  HTTP codes:
    1xx - 0, 2xx - 12157, 3xx - 0, 4xx - 0, 5xx - 0
    others - 0
  Throughput:     1.50MB/s