文件列表:

C10k.png (19778, 2018-09-25)
LICENSE (1068, 2018-09-25)
Manifest.toml (8346, 2018-09-25)
Project.toml (247, 2018-09-25)
results (0, 2018-09-25)
results\apache2.txt (1333, 2018-09-25)
results\julia.txt (1264, 2018-09-25)
results\nodejs.txt (1276, 2018-09-25)
results\plotit.jl (519, 2018-09-25)
src (0, 2018-09-25)
src\C10k.jl (39, 2018-09-25)
src\server.jl (1240, 2018-09-25)
static (0, 2018-09-25)
static\index.html (10831, 2018-09-25)

# C10k *Evaluation of a Julia web server for 10k concurrent connections* ## Introduction While the [Julia](https://julialang.org) language is popular for scientific computing there aren't many examples of it being used for general purpose applications. Here we compare a simple http based file server written in Julia with other alternatives. We run the popular C10k test to see how a Julia based web server handles 10k simultaneous connections. No attempt is made to tune any of the alternatives. The goal is to see how a solution that is quickly put together will perform. ## The Contenders ### Julia 1. [Julia](https://github.com/JuliaLang/julia) v1.0.0 2. [HTTP.jl](https://github.com/JuliaWeb/HTTP.jl) v0.6.14 ### Apache 1. Server version: [Apache](https://httpd.apache.org)/2.4.29 (Ubuntu) 2. Server built: 2018-06-27T17:05:04 ### Node.js based http-server 1. [node.js](https://nodejs.org/en/) v10.11.0 2. [http-server](https://www.npmjs.com/package/http-server) ## Environment ### Hardware 1. CPU: [Intel(R) Core(TM) i7-7500U](https://ark.intel.com/products/95451/Intel-Core-i7-7500U-Processor-4M-Cache-up-to-3_50-GHz-) CPU @ 2.70GHz 2. System Memory: 16 GB 2. Storage: PCIe NVMe Toshiba 512gb SSD ### OS Linux xyxy 4.15.0-33-generic #36-Ubuntu SMP Wed Aug 15 16:00:05 UTC 2018 x86_*** x86_*** x86_*** GNU/Linux ## Benchmark [ab - Apache HTTP server benchmarking tool](https://httpd.apache.org/docs/2.2/programs/ab.html) ```bash bash$ ab -n 13000 -c 10000 -g test.csv http://localhost:8080/index.html ``` ## Julia Web Server A simple file server is implemented using HTTP.jl. ```julia function file_server() web_server = HTTP.Servers.Server(handler, ratelimit=1000000//1) @async HTTP.Servers.serve(web_server, HOST, PORT, false) end function handler(req::HTTP.Request) (req.method != "GET") && return HTTP.Response(404, "Method $(req.method) not supported!") try return HTTP.Response(get_file(req)) catch e return HTTP.Response(404, "Error: $e") end end function get_file(req::HTTP.Request) :: Vector{UInt8} t = (req.target == "/") ? "/index.html" : req.target read(STATIC_DIR * t) end ``` Running the server ```julia (v1.0) pkg> activate . julia> using C10k; C10k.file_server() [ Info: Listening on: Sockets.InetAddr{Sockets.IPv4}(ip"127.0.0.1", 0x1f90) Task (runnable) @0x00007fa9c0730d30 julia> using Logging; global_logger(NullLogger()) # Turn off logging ``` ## Results The benchmark opens 10,000 simultaneous connections to the server and measures the latency of each response. The figure below shows the [percentile](https://www.mathsisfun.com/data/percentiles.html) of requests below a given latency. For example the blue line that shows the Julia server's performance depicts that 68% of the requests completed in less than 100 ms, 75% in less than 1.8 seconds, 80% in less than 3.4 seconds, etc.  * For upto 6.8K simultaneous connections Julia performs 6 times better (below 100 ms in latency per request) than nodejs, roughly equivalent to the native C performance of Apache. * Beyond 6.8K connections Julia's performance steadily degrades. * Nodejs performance degrades more gradually after 6.8K connections. In all Nodejs shows the most consistent performance. * The worst case Latency of the Julia server is 6.7 seconds. * Apache performs very badly on the last connection (possibly last close()). ## Conclusions An HTTP file server in Julia quickly put together with HTTP.jl performs well upto about 5K simultaneous connections. More tests need to be run for heavier workloads that exercise the CPU or memory and the degradation of performance beyond 6.8K connections needs to be studied. ## Appendix ### OS Settings ```bash bash$ sysctl net.ipv4.ip_local_port_range="15000 61000" bash$ sysctl net.ipv4.tcp_fin_timeout=20 bash$ sysctl net.ipv4.tcp_tw_recycle=1 bash$ sysctl net.ipv4.tcp_tw_reuse=1 ``` Increase the user's ulimit in /etc/security/limits.conf like so. user1 hard nofile 100000 user1 soft nofile 100000 Set the user's ulimit in the shell before running the client or server. ```bash ulimit -n 1000000 ```

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：