How Linux creates sockets and counts them
Hey, If you’ve been working with web servers for a little while, you certainly have already hit the classic “address already in use†(EADDRINUSE). Here in this article, we go through not only how to see whether such condition as conditioned to happen (by looking at the list of open sockets), but also verify in the actual Kernel code paths where that check happens. In case you’ve been wondering abo
Scaling techniques for servers with high connection rates
SO_REUSEPORT Scaling Techniques for Servers with High Connection Rates Ying Cai ycai@google.com Problems Servers with high connection/transaction rates TCP servers, e.g. web server UDP servers, e.g. DNS server On multi-core systems, using multiple servicing threads, e.g. one thread per servicing core. The single server socket becomes bottleneck Cache line bounces Hard to achieve load balance Thing
Coping with the TCP TIME-WAIT state on busy Linux servers | Vincent Bernat
TL;DR Do not enable net.ipv4.tcp_tw_recycle—it doesn’t even exist anymore since Linux 4.12. Most of the time, TIME-WAIT sockets are harmless. Otherwise, jump to the summary for the recommended solutions. The Linux kernel documentation is not very helpful about what net.ipv4.tcp_tw_recycle and net.ipv4.tcp_tw_reuse do. This lack of documentation opens the path to numerous tuning guides advising to
Monitoring and Tuning the Linux Networking Stack: Sending Data | Packagecloud Blog
Monitoring and Tuning the Linux Networking Stack: Sending Data TL;DR This blog post explains how computers running the Linux kernel send packets, as well as how to monitor and tune each component of the networking stack as packets flow from user programs to network hardware. This post forms a pair with our previous post Monitoring and Tuning the Linux Networking Stack: Receiving Data. It is imposs
Sendfile (a system call for web developers to know about!)
The other day I learned about a new (to me) exciting Linux system call! (for newcomers, a system call is an operation you can ask the operating system to do). This one seems really important to know about if you’re configuring a webserver! So let’s learn about it. Before this, I knew about basic system calls like open and read for files, and sendto and recvfrom for networking. And a few fancier th
Serializing accept(), AKA Thundering Herd, AKA the Zeeg Problem — uWSGI 2.0 documentation
After having forked itself a bunch of times, each process will generally start blocking on accept() The funny problem is that on older/classic UNIX, accept() is woken up in each process blocked on it whenever a connection is attempted on the socket. Only one of those processes will be able to truly accept the connection, the others will get a boring EAGAIN. This results in a vast number of wasted
Thundering herd 対ç–ã®æœ¬å‘½ã€ EPOLLEXCLUSIVE を試ã—ã¦ã¿ãŸ : DSAS開発者ã®éƒ¨å±‹
epoll を使ã£ãŸ prefork 型アプリケーションサーãƒãƒ¼ã«ãŠã‘ã‚‹ Thundering herd 対ç–ã®æ±ºå®šç‰ˆã¨ã—ã¦æ³¨ç›®ã•ã‚Œã¦ã„㟠EPOLLEXCLUSIVE ãŒã€ 3/13 ã«ãƒªãƒªãƒ¼ã‚¹ã•ã‚ŒãŸ Linux 4.5 ã§å°Žå…¥ã•ã‚Œã¾ã—ãŸã€‚ 昨年 SO_REUSEPORT ã¨ã„ã†ã‚½ã‚±ãƒƒãƒˆã‚ªãƒ—ションãŒç™»å ´ã—ã¦ã€ Thundering herd 対ç–ã¨ã—ã¦è©±é¡Œã«ãªã£ãŸã‚‚ã®ã®ã€ãƒ¯ãƒ¼ã‚«ãƒ¼ã”ã¨ã« listen ã‚ューãŒä½œã‚‰ã‚Œã‚‹ãŸã‚ graceful restart ã™ã‚‹ã¨ãã« listen ã‚ューã«å…¥ã£ã¦ã‚‹ãƒªã‚¯ã‚¨ã‚¹ãƒˆã‚’å–ã‚Šã“ã¼ã™å¯èƒ½æ€§ãŒã‚り利用ã™ã‚‹ã®ãŒé›£ã—ã„状æ³ã§ã—ãŸã€‚ å‚考: epoll ã® thundering herd å•é¡Œã«ã¤ã„ã¦è§£èª¬ã—ã¦ã„るサイト http://tech.geniee.co.jp/entry/so_reuseport http://uwsgi-docs.readthedo
リアルタイム通信環境ã®(一部)構æˆç´¹ä»‹ : DSAS開発者ã®éƒ¨å±‹
ã“ã‚“ã«ã¡ã¯ã€‚ 今回ã¯ã€å½“社ã§ç¨¼åƒã•ã›ã¦ã„るリアルタイム通信環境ã«ã¤ã„ã¦ã€ã”紹介ã•ã›ã¦é ‚ãã¾ã™ã€‚ ã”紹介ã™ã‚‹ç’°å¢ƒã«å¯¾ã™ã‚‹è¦ä»¶ã¯ã€ä»¥ä¸‹ã¨ãªã‚Šã¾ã™ã€‚ ・ゲーム内ã®æœŸé–“é™å®šã‚¤ãƒ™ãƒ³ãƒˆã§ä½¿ç”¨ã—ã€ã‚¤ãƒ™ãƒ³ãƒˆé–‹å‚¬ä¸ã®ã¿ã‚µãƒ¼ãƒã‚’ç¨¼åƒ ãƒ»ãƒªã‚¢ãƒ«ã‚¿ã‚¤ãƒ é€šä¿¡ã€‚ãƒ—ãƒãƒˆã‚³ãƒ«ã¯ã€websocket を使用 ・åŒã˜ãƒãƒ¼ãƒ ã«æ‰€å±žã™ã‚‹ãƒ¦ãƒ¼ã‚¶ã‚’åŒã˜ã‚µãƒ¼ãƒã¸æŽ¥ç¶š ・ã¨ã‚Šã‚ãˆãšã„ã£ã±ã„スケールã§ãるよã†ã«(笑 最後ã®è¦ä»¶ã¯å†—談ã§ã€å®Ÿéš›ã«ã¯ã¡ã‚ƒã‚“ã¨ã—ãŸæ•°å€¤ã‚’é ‚ã„ã¦ã„ã‚‹ã®ã§ã™ãŒã€ã“ã®ã‚ˆã†ãªç’°å¢ƒæ§‹ç¯‰ã‚’ä¾é ¼ã•ã‚Œã¾ã—ãŸã®ã§ã€AWS 上ã§ä»¥ä¸‹ã«ã‚るよã†ãªæ§‹æˆã‚’考ãˆã¦ã¿ã¾ã—ãŸã€‚ 構æˆå›³ ※ 主è¦ãªã‚µãƒ¼ãƒã®ã¿ã‚’抜粋 ELB 外部ã®ã‚¯ãƒ©ã‚¤ã‚¢ãƒ³ãƒˆã‹ã‚‰ã€websocket ãªæŽ¥ç¶šã‚’å—ã‘付ã‘ã¾ã™ã€‚ http(s) モードã§ã¯ã€websocket ã®é€šä¿¡ç¢ºç«‹ã«å¿…è¦ãªãƒ˜ãƒƒãƒ€ãŒæ¶ˆåŽ»ã•ã‚Œã¦ã—ã¾ã†ãŸã‚ã€tcp モードを使用ã—ã¦ã„ã¾ã™ã€‚ tcp モードを有効ã«ã™ã‚‹ã¨ã€
How to receive a million packets per second
Last week during a casual conversation I overheard a colleague saying: "The Linux network stack is slow! You can't expect it to do more than 50 thousand packets per second per core!" That got me thinking. While I agree that 50kpps per core is probably the limit for any practical application, what is the Linux networking stack capable of? Let's rephrase that to make it more fun: On Linux, how hard
prefork サーãƒãƒ¼ã¨ thundering herd å•é¡Œ - naoyaã®ã¯ã¦ãªãƒ€ã‚¤ã‚¢ãƒªãƒ¼
Catalyst ã‚’ POE ã§å‹•ã‹ã™ Engine ã® Catalyst::Engine::HTTP::POE ã¨ã„ã†å®Ÿè£…㌠CPAN ã«ã‚ã‚Šã¾ã™ã€‚"Single-threaded multi-tasking Catalyst engine " ã ãã†ã§ã™ã€‚"Single-threaded" ã¨è¨€ã„ã¤ã¤ã‚‚実装を覗ã„ã¦ã¿ã‚‹ã¨ç’°å¢ƒå¤‰æ•° CATALYST_POE_MAX_PROC ã‚’ 1 よりも大ããè¨å®šã™ã‚‹ã“ã¨ã§ prefork ã™ã‚‹å®Ÿè£…ã«ãªã£ã¦ã¾ã™ã€‚POEシングルスレッドã§ã¯ã‚¢ãƒ—リケーション内ã§ç™ºç”Ÿã™ã‚‹ãƒ–ãƒãƒƒã‚¯ã‚’é¿ã‘ã‚‹ã“ã¨ãŒé›£ã—ã„ã®ã§ãã®ãŸã‚ã®å®Ÿè£…ã˜ã‚ƒãªã„ã‹ãªã¨æ€ã„ã¾ã™ã€‚ ã¨ã“ã‚ã§ã“ã® Catalyst POE エンジンã€prefork ã®å®Ÿè£…ã¯ã©ã®ã‚ˆã†ã«è¡Œã£ã¦ã„ã‚‹ã‹ã¨ã„ã†ã¨ POE ã‹ã‚‰ prefork ã¨åã®ä»˜ã„ãŸã‚¤ãƒ™ãƒ³ãƒˆãŒç™ºç”Ÿã™ã‚‹ã¨ãŠã‚‚ã‚€ã‚ã«åプãƒã‚»ã‚¹ã‚’生æˆã™ã‚‹ã€ã¨ã„ã†ã®ã‚‚ã®ã€‚複数ã®
The Implementation of epoll (1)
In this series of blog posts, I will be taking notes about the internals and implementation details of the Linux event poll (epoll) subsystem. Prerequisites Please note that I am assuming the audiences of those articles are familiar with epoll APIs and usages. I will be focusing on the actual kernel implementation of the epoll subsystem, not it's usages. If you are not familiar with the usage of e
Socket Sharding in NGINX Release 1.9.1
F5 Sites DevCentral Connect & learn in our hosted community F5 Labs The latest threat intel and research to help protect your apps MyF5 Your key to everything F5, including support, registration keys, and subscriptions Partner Central Research and support for partners LearnF5 Guidance, insights, and how to use F5 products Contact F5 Contact F5 Sales Talk to an F5 sales representative Contact F5 Su
UNIXã®ãƒ—ãƒã‚»ã‚¹ã‚„シグナルをã—ã£ã‹ã‚Šç†è§£ã™ã‚‹ãŸã‚ã®æŠ€è¡“ドã‚ュメント「Process Book〠| ソフトアンテナ
UNIXã®åŸºæœ¬ã‚’ãªã™ãƒ—ãƒã‚»ã‚¹ã‚„シグナルãªã©ã‚’分ã‹ã‚Šã‚„ã™ã解説ã—ãŸæŠ€è¡“ドã‚ュメント「Process Bookã€ãŒGitHubã«ã¦å…¬é–‹ã•ã‚Œã¦ã„ã¾ã™ã€‚ åŒãƒ‰ã‚ュメントã¯ã€ãƒ—ãƒã‚»ã‚¹ã®ç”Ÿæˆã€ãƒ—ãƒã‚»ã‚¹ã¨ãƒ•ã‚¡ã‚¤ãƒ«å…¥å‡ºåŠ›ã€ãƒ•ã‚¡ã‚¤ãƒ«ãƒ‡ã‚£ã‚¹ã‚¯ãƒªãƒ—ã‚¿ã€preforkサーãƒãƒ¼ã®ä½œã‚Šæ–¹ã€ã‚¾ãƒ³ãƒ“プãƒã‚»ã‚¹ã¨å¤å…プãƒã‚»ã‚¹ã€ã‚·ã‚°ãƒŠãƒ«ã¨killã€ãƒ—ãƒã‚»ã‚¹ã‚°ãƒ«ãƒ¼ãƒ—ã¨ãƒ•ã‚©ã‚¢ã‚°ãƒ©ãƒ³ãƒ‰ãƒ—ãƒã‚»ã‚¹ã¨ã„ã£ãŸè©±é¡Œã‚’解説ã™ã‚‹ãƒ‰ã‚ュメントã§ã€GitHubã®releasesディレクトリã«ã¯PDFã€epubå½¢å¼ã®ãƒ•ã‚¡ã‚¤ãƒ«ã‚‚æ ¼ç´ã•ã‚Œã¦ã„ã¾ã™ã€‚ UNIX環境ã§ãƒ—ãƒã‚°ãƒ©ãƒ を開発ã™ã‚‹éš›ã«å½¹ç«‹ã¤ã ã‘ã§ã¯ãªãã€æ™®é€šã®ãƒ¦ãƒ¼ã‚¶ãƒ¼ã¨ã—ã¦ã‚¿ãƒ¼ãƒŸãƒŠãƒ«ã‚’使ã£ã¦ã‚³ãƒžãƒ³ãƒ‰æ“作を行ã†éš›ã«ã‚‚å½¹ã«ç«‹ã¡ãã†ãªå†…容ã ã¨æ€ã„ã¾ã™ã€‚
Epoll - from the kernel side
The document discusses the basics of I/O and event-driven I/O models in Linux like blocking I/O, non-blocking I/O, I/O multiplexing using select/poll and their internals. It then introduces epoll as a more efficient alternative to select/poll and describes its user-space API, kernel structures and how it works by adding file descriptors to an epoll instance and waking up blocked processes on I/O e
somaxconnã®ä¸Šé™å€¤ã¯65535
Linuxã®ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ãƒ‘ラメータã®ä¸€ã¤ã«ã€net.core.somaxconnã¨ã„ã†ã®ãŒã‚ã‚Šã¾ã™ã€‚ã“ã‚Œã¯listen(2)ã®ç¬¬äºŒå¼•æ•°backlogã®ä¸Šé™å€¤ã¨ãªã£ã¦ã„ã¾ã™ã€‚ã“ã®somaxconnã¯ä¸€è¦‹intã«è¦‹ãˆã¾ã™ãŒã€å®Ÿã¯unsigned shortã®ç¯„囲ã®æ•°å€¤ã—ã‹å—ã‘付ã‘ã¾ã›ã‚“。ãれを超ãˆã‚‹æ•°å€¤ã‚’入れるã¨é»™ã£ã¦åˆ‡ã‚Šæ¨ã¦ã‚‰ã‚Œã¾ã™ã€‚ã¤ã¾ã‚Š -1→65535 0→0 65535→65535 65536→0 65537→1 ã¨åŒã˜ã‚ˆã†ãªå‹•ä½œã‚’内部的ã«ã—ã¾ã™ã€‚ãªã®ã§ã€ã“ã®å€¤ã¯çµ¶å¯¾ã«0～65535ã®ç¯„囲を超ãˆã¦ã¯ã„ã‘ã¾ã›ã‚“。以下ã€è©³ã—ã„説明ã§ã™ã€‚ ãŠã“ã¨ã‚ã‚Š: ã“ã®ä»•æ§˜ã¯Linux 3.11以é™å¤‰æ›´ã•ã‚Œã¦ãŠã‚Šã€ç¯„囲外ã®æ•°å€¤ã‚’è¨å®šã§ããªã„よã†ã«ãªã£ã¦ã„ã¾ã™ã€‚ã“ã“ã«æ›¸ã„ã¦ã‚る内容ãŒå†ç¾ã™ã‚‹ã®ã¯ã€Linux 3.10以å‰ã®å¤ã„カーãƒãƒ«ã®ã¿ã§ã™ã€‚ ã¾ãšsysctlã®å®šç¾©ã§ã™ãŒ net/core/sysct
socat を使ã†ã¨å„種ソケットã®æ“作ãŒæ—ã‚Šã¾ãる件
socat ã£ã¦ã„ã†ãƒ„ールを使ã£ã¦ã¿ãŸã¨ã“ã‚ソケット関連ã®ã¡ã‚‡ã£ã¨ã—ãŸæ“作をã“ãªã™ã®ã«ã™ã“ã¶ã‚‹ä½¿ã„å‹æ‰‹ãŒè‰¯ã‹ã£ãŸã€‚ インストール Mac OS X ãªã‚‰ Homebrew ã§ã‚¤ãƒ³ã‚¹ãƒˆãƒ¼ãƒ«ã§ãる。 $ brew install socat $ which socat /usr/local/bin/socat Echo サーãƒã‚’ç«‹ã¦ã‚‹ socat コマンドã«ã¯ã‚½ã‚±ãƒƒãƒˆã®å…¥åŠ›ã¨å‡ºåŠ›ã‚’指定ã™ã‚‹ã€‚ 以下ã®ã‚ˆã†ã«ã™ã‚‹ã¨ TCP 8080 ãƒãƒ¼ãƒˆã§æŽ¥ç¶šã‚’å¾…ã¡å—ã‘ã¦å‡ºåŠ›ã¯å…¥åŠ›ã‚’ãã®ã¾ã¾ cat ã™ã‚‹ã“ã¨ã§ TCP ã® Echo サーãƒã«ãªã‚‹ã€‚ v オプションã§é€šä¿¡å†…容を表示ã™ã‚‹ã€‚ $ socat -v tcp-listen:8080 system:cat 別ã®ã‚¿ãƒ¼ãƒŸãƒŠãƒ«ã‹ã‚‰ telnet コマンドã§æŽ¥ç¶šã—ã¦ã¿ã‚‹ã€‚ $ telnet localhost 8080 Trying ::1... telnet:
HttpClientã¨ConnectionManagerã¨CLOSE_WAIT㨠- CLOVERðŸ€
Javaã§HTTP通信をやã‚ã†ã¨ã™ã‚‹ã¨ã€å¿…ãšã¨è¨€ã£ã¦ã„ã„ã»ã©ç™»å ´ã™ã‚‹ã€Apache HttpClient4（以å‰ã¯ã€Jakarta Commons HttpClient3）ã§ã™ãŒã€ã“ã‚Œã§ã¡ã‚‡ã£ã¨ãƒãƒžã£ã¦ã„ãŸäº‹è±¡ã‚’見ã‹ã‘ãŸã®ã§ã€è»½ãメモを。 ã‚ã€ãã®æ™‚ã«ä½¿ã‚ã‚Œã¦ã„ãŸã®ã¯ã€Jakarta Commons HttpClient3ã®æ–¹ã§ã™ã€‚ã‚“ã§ã€ã‚½ã‚±ãƒƒãƒˆã®ã‚¯ãƒãƒ¼ã‚ºæ¼ã‚Œã§ãƒãƒžã£ã¦ã¾ã—ãŸã€‚ ã“ã‚Œã€ãƒãƒ¥ãƒ¼ãƒˆãƒªã‚¢ãƒ«ã ã‘ã‚’ä¿¡ã˜ã¦å®Ÿè£…ã™ã‚‹ã¨ã€ãƒãƒžã‚‹ã‚“ã§ã™ã‚ˆãã‡â€¦ã€‚ http://hc.apache.org/httpclient-3.x/tutorial.html ãƒãƒ¥ãƒ¼ãƒˆãƒªã‚¢ãƒ«ã®ä¸ã« // Release the connection. method.releaseConnection(); ã¿ãŸã„ãªã“ã¨ãŒæ›¸ã„ã¦ã‚ã£ã¦ã€ã„ã‹ã«ã‚‚コãƒã‚¯ã‚·ãƒ§ãƒ³ã‚’クãƒãƒ¼ã‚ºã—ã¦ãã‚Œãã†ãªé›°å›²æ°—ãŒã‚ã‚Šã¾ã™ãŒã€ã“ã‚Œã¯æ„味ãŒé•ã„ã¾ã™ã€‚Ht
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
GitHub - google/tcp_killer: Shuts down a TCP connection on Linux or macOS. Local and remote endpoint arguments can be copied from the output of 'netstat -lanW'.
How TCP Sockets Work
Reducing Network Latency - Jesse Brandeburg : Blueprints : Linux Plumbers Conference
{{#tags}}- {{label}}
{{/tags}}