Amazon EC2 的網路效能

前一篇「在 AWS 上面的 OpenVPN Server 效能」最後的問題就是 EC2 instance 本身的網路效能,畢竟是公司要用的,還是實際測一下數字,之後有人接手的時候也比較清楚是怎麼選這個大小的...

這邊拿的是 AWSap-southeast-1 (Singapore) 的 EC2 測試,直接在同一個 subnet 裡面開兩台一樣的機器跑 iperf 測試。

機器開機後會先跑這串指令 (除了安裝 iperf 的指令,其他的是出自我自己 wiki 上的 Ubuntu 這頁),然後再重開機:

sudo fallocate -l 512M /swapfile; sudo chmod 600 /swapfile; sudo mkswap /swapfile; sudo swapon /swapfile; echo '/swapfile none swap sw 0 0' | sudo tee -a /etc/fstab; echo -e "net.core.default_qdisc=fq\nnet.ipv4.tcp_congestion_control=bbr" | sudo tee /etc/sysctl.d/99-tcp.conf; sudo sysctl -p /etc/sysctl.d/99-tcp.conf; sudo apt update; sudo apt dist-upgrade -y; sudo apt install -y apache2-utils apt-transport-https build-essential curl dnsutils dstat git jq locales moreutils most mtr-tiny net-tools p7zip-full pigz prometheus-node-exporter rsync sharutils software-properties-common sysstat unrar unzip vim-nox wget zsh zsh-syntax-highlighting zstd; sudo apt install -y iperf; sudo apt clean

接下來就是一台跑 iperf -s,另外一台跑 iperf -c 10.x.x.x -i 1 -t 3600 讓他跑一個小時看結果了。

我都有跑 tmux 再連到這些機器上,這樣可以捲回去看每一秒的傳輸速度,就可以看出來變化了,不過這邊還是簡單的只列出最高速度 (burstable) 與穩定輸出的速度 (baseline):

EC2 instance Baseline Burstable vCPU RAM Pricing (USD$)
c6g.medium 500Mbps 10Gbps 1 2GB 0.0392
c6g.large 750Mbps 5Gbps (claimed 10Gbps) 2 4GB 0.0784
c6g.xlarge 1.25Gbps 10Gbps 4 8GB 0.1568
t4g.small 125Mbps 5Gbps 2 2GB 0.0212
t4g.medium 255Mbps 5Gbps 2 4GB 0.0424
t4g.large 510Mbps 5Gbps 2 8GB 0.0848
t4g.xlarge 1Gbps 5Gbps 4 16GB 0.1696

這邊沒列出來的是 burstable 可以持續的時間,但這跟你機器吃的網路資源有關,我就決定只用 baseline 來做決策了,這樣可能會多花一點錢,但會少很多麻煩。

另外這次在處理的過程有被同事提醒各種 bandwidth overhead,所以就順便查了一下資料:

  • OpenVPN 本身的 overhead 大約是 5% (跑 UDP 的時候):「OpenVPN performance」。
  • SSH 也有些 overhead,大約是 6% (把來回的封包都算進去):「What is the overhead of SSH compared to telnet?」。
  • rsync 的部份鐵定也有 overhead,但這邊就沒找到現成的文章有統計過了。
  • 另外我自己之前做實驗發現 TCP BBR 的 retransmission algorithm 還蠻激進的,會有 10% packet loss,改用預設的 CUBIC 會好很多,大約 1% 到 2% 左右。

綜合這些測試,我自己抓了 35% 的 overhead 來推估,最後是用 c6g.large 來養 VPN server。750Mbps 的實際流量大約可以包進 550Mbps 的原始流量,大約是 68MB/sec。

不過新加坡與印尼之間的 internet bandwidth 好像還是不太夠,有時候深夜跑也跑不滿... 不過之後 VPN 上的 client 會愈來愈多,應該是不需要降...

JavaScript Framework 不可避免的成本

看到「The Baseline Costs of JavaScript Frameworks」這篇文章在研究目前主流 JavaScript Framework 無法避免的成本到底有多高。

文章的結論是目前常見的 JavaScript Framework 其實都很肥重,在網路速度不快的地方得花不少時間下載,在非旗艦的手機上會需要花不少時間處理 (parse & compile)。

這是 gzip 後的大小:

這是 parse & compile 的時間:

這是下載時間 (扣除 latency 與 TLS connection 建立時間):

並不是說不能用,但重點會在客群:

But it’s important to consider your audience. If you’re building for resource constrained devices — which you certainly are if your product targets a country like India — you could consider using a lighter framework such as Riot or Preact. Your users will thank you.

最後有建議如果只是要呈現資訊,不要用整套 JavaScript Framework,在有需要互動的地方另外寫就好了:

For websites that primarily display content, it’s more efficient and cost-effective to just send some server-rendered HTML down the wire. If there are areas of your website that require interactivity, you can always use JavaScript to build those specific parts.

不同性質的應用程式對 KPTI (Meltdown 修正) 的效能影響

NetflixBrendan Gregg 整理了他測試 KPTI 對效能的影響:「KPTI/KAISER Meltdown Initial Performance Regressions」。

與其他人只是概括的測試,他主要是想要針對可量測的數字對應出可能的 overhead,這樣一來還沒上 patch 的人就可以利用這些量測數字猜測可能的效能衝擊。

他把結論放在前面:

To understand the KPTI overhead, there are at least five factors at play. In summary:

  • Syscall rate: there are overheads relative to the syscall rate, although high rates are needed for this to be noticable. At 50k syscalls/sec per CPU the overhead may be 2%, and climbs as the syscall rate increases. At my employer (Netflix), high rates are unusual in cloud, with some exceptions (databases).
  • Context switches: these add overheads similar to the syscall rate, and I think the context switch rate can simply be added to the syscall rate for the following estimations.
  • Page fault rate: adds a little more overhead as well, for high rates.
  • Working set size (hot data): more than 10 Mbytes will cost additional overhead due to TLB flushing. This can turn a 1% overhead (syscall cycles alone) into a 7% overhead. This overhead can be reduced by A) pcid, available in Linux 4.14, and B) Huge pages.
  • Cache access pattern: the overheads are exacerbated by certain access patterns that switch from caching well to caching a little less well. Worst case, this can add an additional 10% overhead, taking (say) the 7% overhead to 17%.

重點在於給了量測的方式,以第一個 Syscall rate 來說好了,他用 sudo perf stat -e raw_syscalls:sys_enter -a -I 1000 測試而得到程式的 syscall 數量,然後得到下面的表格,其中 X 軸是每秒千次呼叫數,Y 軸是效能損失:

用這樣的方式提供給整個組織 (i.e. Netflix) 內評估衝擊。

zetcd:用 etcd 跑 ZooKeeper 架構

在「zetcd: running ZooKeeper apps without ZooKeeper」這邊介紹了用 etcd 當作 ZooKeeper 伺服器。程式碼在「Serve the Apache Zookeeper API but back it with an etcd cluster」這邊可以看到。

不過可以看到有不少 overhead:

但 etcd 用 Go 寫 (省下 JVM tuning?),可能是個不錯的誘因...

印 "#" 比印 "B" 來的快的問題

這篇是兩年前在 StackOverflow 上的問題:「Why is printing “B” dramatically slower than printing “#”?」。

問問題的人這段程式跑了 8.52 秒:

Random r = new Random();
for (int i = 0; i < 1000; i++) {
    for (int j = 0; j < 1000; j++) {
        if(r.nextInt(4) == 0) {
            System.out.print("O");
        } else {
            System.out.print("#");
        }
    }

   System.out.println("");
 }

而把上面的 # 換成 B 就變成 259.152 秒。

答案是與 word-wrapping 有關:

Pure speculation is that you're using a terminal that attempts to do word-wrapping rather than character-wrapping, and treats B as a word character but # as a non-word character. So when it reaches the end of a line and searches for a place to break the line, it sees a # almost immediately and happily breaks there; whereas with the B, it has to keep searching for longer, and may have more text to wrap (which may be expensive on some terminals, e.g., outputting backspaces, then outputting spaces to overwrite the letters being wrapped).

But that's pure speculation.

這真是細節 XDDD