处理 JSON 的性能表现

以下对 Z-Data 中的 JSON 模块和 Hackage 上被普遍使用的 JSON 处理软件包 aeson进行了基准性能分析。我们的分析结果可以参见这篇博客。

分析结果

处理 TCP 的性能表现

以下通过使用 Redis 软件包中的工具 redis-benchmark 运行 redis-benchmark -p 8888 -t ping -n 100000 -q -c 100 来对不同 Redis 乒乓服务器实现进行了基准性能分析。

# Haskell 的 [network](https://hackage.haskell.org/package/network) 软件包
# cabal run redis-benchmark-base -- +RTS -N4 -H2G
PING_INLINE: 88105.73 requests per second
PING_BULK: 87873.46 requests per second

# Z.Haskell 的 Z-IO 软件包
# cabal run redis-benchmark-z -- +RTS -N4 -H2G
PING_INLINE: 99800.40 requests per second
PING_BULK: 102459.02 requests per second

# Golang 标准库
PING_INLINE: 98716.68 requests per second
PING_BULK: 101522.84 requests per second

# Rust mio
PING_INLINE: 111731.84 requests per second
PING_BULK: 112612.61 requests per second

# C libuv
PING_INLINE: 109170.30 requests per second
PING_BULK: 105374.08 requests per second

注意到 Rust 的 mio 和 C 的 libuv 都使用了单线程的事件循环和一个共享的缓冲区去接收来自 redis-benchmark 不同连接间的消息，和其它基于轻量线程的实现有较大不同。

以下是 GHC 提供的有关内存数据：

# Haskell 的 [network](https://hackage.haskell.org/package/network) 软件包
# cabal run redis-benchmark-base -- +RTS -N4 -s -H2G
 3,751,313,096 bytes allocated in the heap
     302,793,568 bytes copied during GC
       1,869,864 bytes maximum residency (1044 sample(s))
         490,016 bytes maximum slop
            2085 MiB total memory in use (0 MB lost due to fragmentation)

                                     Tot time (elapsed)  Avg pause  Max pause
  Gen  0      2085 colls,  2085 par    0.455s   0.131s     0.0001s    0.0085s
  Gen  1      1044 colls,  1043 par    0.419s   0.149s     0.0001s    0.0226s

  Parallel GC work balance: 82.10% (serial 0%, perfect 100%)

  TASKS: 10 (1 bound, 9 peak workers (9 total), using -N4)

  SPARKS: 0 (0 converted, 0 overflowed, 0 dud, 0 GC'd, 0 fizzled)

  INIT    time    0.001s  (  0.001s elapsed)
  MUT     time    5.360s  (  5.842s elapsed)
  GC      time    0.874s  (  0.280s elapsed)
  EXIT    time    0.001s  (  0.008s elapsed)
  Total   time    6.236s  (  6.130s elapsed)

  Alloc rate    699,915,737 bytes per MUT second

  Productivity  85.9% of total user, 95.3% of total elapsed

# Z.Haskell 的 Z-IO 软件包
# cabal run redis-benchmark-z -- +RTS -N4 -s -H2G
 280,828,448 bytes allocated in the heap
         835,688 bytes copied during GC
       3,375,112 bytes maximum residency (4 sample(s))
         839,672 bytes maximum slop
            2084 MiB total memory in use (0 MB lost due to fragmentation)

                                     Tot time (elapsed)  Avg pause  Max pause
  Gen  0         5 colls,     5 par    0.009s   0.008s     0.0015s    0.0073s
  Gen  1         4 colls,     3 par    0.023s   0.021s     0.0052s    0.0194s

  Parallel GC work balance: 82.81% (serial 0%, perfect 100%)

  TASKS: 14 (1 bound, 13 peak workers (13 total), using -N4)

  SPARKS: 0 (0 converted, 0 overflowed, 0 dud, 0 GC'd, 0 fizzled)

  INIT    time    0.001s  (  0.001s elapsed)
  MUT     time    2.811s  (  6.757s elapsed)
  GC      time    0.032s  (  0.028s elapsed)
  EXIT    time    0.002s  (  0.004s elapsed)
  Total   time    2.846s  (  6.790s elapsed)

  Alloc rate    99,903,441 bytes per MUT second

  Productivity  98.8% of total user, 99.5% of total elapsed

目录

处理 JSON 的性能表现

处理 TCP 的性能表现