另外是套件的部分，照這個文字的說明，應該不是所有的套件都可以丟進去用 (can import a subnet of popular Python packages)，支援的套件看起來是預先 compile 好：
All bindings, including bindings to Vectorize, Workers AI, R2, Durable Objects, and more are supported on day one. Python Workers can import a subset of popular Python packages including FastAPI, Langchain, Numpy and more. There are no extra build steps or external toolchains.
The goal of Facebook’s SSL bump technology was the company’s acquisition, decryption, transfer, and use in competitive decisionmaking of private, encrypted in-app analytics from the Snapchat, YouTube, and Amazon apps, which were supposed to be transmitted over a secure connection between those respective apps and secure servers (sc-analytics.appspot.com for Snapchat, s.youtube.com and youtubei.googleapis.com for YouTube, and *.amazon.com for Amazon). Id.
Today we are using the Onavo vpn-proxy stack to deploy squid with ssl bump the stack runs in edge on our own hosts (onavopp and onavolb) with a really old version of squid (3.1).
This list is a minimum. Some servers have accelerators and you'll want their analysis tools installed as well: e.g., on Intel GPU servers, the intel-gpu-tools package; on NVIDIA, nvidia-smi. Debugging tools, like gdb(1), can also be pre-installed for immediate use in a crisis.
Garnet is a remote cache-store from Microsoft Research that offers strong performance (throughput and latency), scalability, storage, recovery, cluster sharding, key migration, and replication features. Garnet can work with existing Redis clients.
The magic of this algorithm is that you no longer need to use an amount of memory proportional to the number of items counted, and instead can use a constant amount of memory; 12k bytes in the worst case, or a lot less if your HyperLogLog (We'll just call them HLL from now) has seen very few elements.
在檔案開頭的註解可以看到有 16384 個 register (對應到論文裡面的 b = 14，因為 214 = 16384)，單個 register 的大小則是 6 bit (對應到論文裡面的 M[j])，相乘後是 12K bytes，剛好符合文件上的說明：
The use of 16384 6-bit registers for a great level of accuracy, using a total of 12k per key.
在「Dense representation」這邊也說明了每個 register 都是 6 bit 的存放方式，到這邊都與 HLL 論文提到的實作一樣。
省空間的方式是在「Sparse representation」這邊做到的，在大多數的 register 都沒有被設定的情況下，用這種方式可以省下大量的空間，而缺點是當元素「有點多」的時候會有比較高的 CPU time：
In the example the sparse representation used just 7 bytes instead of 12k in order to represent the HLL registers. In general for low cardinality there is a big win in terms of space efficiency, traded with CPU time since the sparse representation is slower to access.
Cross-platform GUI written in Rust using ADB to debloat non-rooted android devices. Improve your privacy, the security and battery life of your device.
In both our GRPC and DGS Framework services, GC pauses are a significant source of tail latencies.
That’s particularly true of our GRPC clients and servers, where request cancellations due to timeouts interact with reliability features such as retries, hedging and fallbacks.
ZGC has a fixed overhead 3% of the heap size, requiring more native memory than G1. Except in a couple of cases, there’s been no need to lower the maximum heap size to allow for more headroom, and those were services with greater than average native memory needs.
第四張則是 Huge Pages 的差異，這邊要注意這張圖的 Y 軸不是從 0 計算：
可以看到在開 Huge Pages 後，在 RPS (request per second) 不變的情況下 CPU 使用率是有下降的，大約從 50% 降到 45% 左右，不過這張圖的時間跨度有點少，應該是要拉長一點的圖... 不過既然被提出來了，就假設 Netflix 內看起來應該是有這個趨勢，只是抓圖的時候懶了點？
