latency on S Anand

Things I Learned - 14 Sep 2025

Sun, 14 Sep 2025 00:00:00 +0000

This week, I learned:

Though I’m connected on LinkedIn with people I can’t remember (weak ties), pruning them shrinks serendipity. Weak ties, despite noise, are disproportionately valuable for opportunities, e.g. intros, jobs, and pruning reduces future upside. Science
Claude has a Python + Node code interpreter that can access GitHub, PyPi, npm and Google. Simon Willison
SuperTinyIcons has very small icons for many websites and is available via CDN. Sample: http://cdn.jsdelivr.net/npm/super-tiny-icons/images/svg/github.svg
Clock bench is an LLM benchmark based on how well LLMs tell the time from an analog clock. Humans (89%) are much better than the best model (Gemini 2.5 Pro - 13%).
Veo 3 is now available via API. Veo 3 fast is 15s/second. Google
ChatGPT has full support for MCPs via Developer mode in Plus and Pro accounts, via “Developer mode”. OpenAI
In Pyodide, you can use from js import document and then document.querySelector to manipulate the DOM directly from Python. from pyodide.http import pyfetch lets you use fetch.
gtrending is a Python package that fetches trending GitHub repos, users, etc. uvx gtrending repos --language rust --since weekly fetches trending Rust repos of the week.
astgrep lets you search in code (across languages) using AST patterns. Like semgrep but more about code search than security. uvx --from ast-grep-cli ast-grep runs from the CLI. Useful for code rewriting, fast linting, code search.
hurl is a CLI config-based HTTP automation tool. Useful for tests, bulk (templatized) HTTP requests, etc.
rustdesk is an open-source remote desktop software. TeamViewer alternative. Self-hostable.
prek is a much faster version of pre-commit - a cross-language pre-commit hook manager.
⭐ mise is a tool version manager. Combines nvm/fnm, pipx, etc. Supports running several tools with a smooth installation.
The npm phishing email was a great one. It compromised chalk which is used in most npm packages. This may be one of the best supply chain attacks in recent times and makes me want to pin versions instead of using npx -y. Also makes me glad that I’m sponsoring @isaacs and @sindresorhus - two critical open source maintainers.
“I pay for YouTube Premium. For my money, it’s the best bang-for-the-buck subscription service on the market”. - Gavin Andregg
LLMs are non deterministic because GPUs add floating point numbers concurrently and FP addition is non associative - order matters. Thinking Machines
Claude.ai can natively work with Excel, PPTX, DOCX, and PDF files now.
With embeddings, atomic labels + hierarchy beat instruction-heavy prompts. Prefer short, concrete sub-labels (e.g., “promotion,” “job security,” “flexibility”) that roll up to a parent “career” rather than a composite instruction like “Total Rewards and Career Growth”. Embedding similarity is not smart enough to figure this out.
Today, RPA is cheaper than LLMs in some areas. But it’s a moving target. LLM costs are fall fast: 70–90% declines across major providers in 1.5 years. Therefore, waiting has option value. But classic IT compares static quotes, not declining curves, and hence is likely to under-procure LLM solutions.
⭐ The biggest near-term ROI for LLMs in data science is like ‘boring’ data work: PII tagging, data dictionaries, ER/joins, SDTM mapping, etc.. People expect flashy GenAI, but LLMs can bootstrap schema matching and data-cleaning, speeding engineer verification, which is more useful at scale.
You can create an infinite leaflet map with nano banana.
Codex CLI with high reasoning effort seems far more comprehensive than Codex online. I asked both to identify the system requirements (URLs to access, software to install, ports to open) for my Tools in Data Science course. Codex CLI got it right one shot (after 10 minutes of thinking). Codex online missed several items even after 4 attempts.
The Reod on Elantris might have been triggered by Jaddeth who might be an Autonomy avatar. ChatGPT
Output tokens dominate latency. Decoding is sequential (one token depends on all prior tokens), so long completions are the main throttle. Shrinking returned text (e.g., send spans/tags instead of echoing paragraphs) yields a far bigger win on latency than shrinking inputs.

Hosting options

Sat, 01 Jun 2013 05:48:55 +0000

I've been trying out a number of options for hosting recently, and have settled on Amazon spot instances.

Here were my options:

Application hosting, like Google AppEngine. I used this a lot until 2 years ago. Then they changed their pricing, and I realised what “lock-in” means. I can’t just take that code and move it to another server. Besides, I’m a bit wary of Google pulling the plug. Heroku? Same problem. I just want to take the code elsewhere and run it.
Shared hosting, like Hostgator. This blog is run on Hostgator and I’m extremely happy with them. But the trouble is, with shared hosting, I don’t get to run long-running processes on any ports I like.
Run you own servers. The problem here is quite simple: power cuts in India.
Dedicated hosting, like Amazon EC2, Azure, GCE, etc. This remains as pretty much the main hosting option

I’m a price optimisation freak. So I ran the numbers for a year’s worth of usage. I was looking at the CPU cost of a large machine with 7-8GB RAM. Bandwidth and storage are negligible. The cost per hour worked out to:

Amazon: $0.32 / hr in Singapore, $0.24 in Virginia
Google: $0.29 / hr in Europe
Microsoft: $0.32 / hr in US

The price is not all that different, but I need low latency, so Singapore it what it’ll have to be.

EC2 location	Latency (ms)
Singapore	139
Oregon, US	334
Japan	517
Ireland	618
Australia	620
California, US	677
Virginia, US	710

Now comes the choice of the right model. At $0.32 per hour, that’s $230 a month.

Amazon offers some ways of getting this down. Instead of on-demand instances, I could go for reserved instances. For a year of usage, that’d get the price down to about $131 a month, nearly halving it. ($739 upfront for a heavy utilisation large reserved instance, with $0.095 * 24 * 365.25 for the year.)

In this case, I know I’ll need the servers for a year. Probably more, but then, I might want to switch later. So this isn’t a bad move. But we can do better. Amazon also offers spot instances. Spot instances might get shut down any time – but in reality, so can on-demand instances. I need to plan for it anyway. I’m not going to host anything that’s so sensitive that if it’s down for a few hours, I’ll have a problem.

But what’s attractive is the pricing. Typically, it’s $0.04 per hour, making it about $29 per month. Even if it shoots up to twice that, at $58, it’s less than a fourth of the on-demand price and less than half the reserved instance price.

I’ve managed to script the entire setup up sequence as shell scripts, and it takes less than an hour to get a new server up and running the software I need. I need to work out a decent backup mechanism. Plus, I could use more reliable storage like like Amazon’s EBS to preserve the data. But on the whole, the pricing is far too attractive and makes the risks worthwhile.

Comments

Anand Chitipothu 2 Jun 2013 3:36 pm: Did you look at Linode? I’ve been using Linode for couple of years and pretty happy with it. The 8GB RAM model costs about $160 per month. Sadly, they don’t have a data center in Singapore. You can test the download speeds and latency of their data centers from https://www.linode.com/speedtest/.
jouko 7 Jul 2013 4:21 pm: I am using Inmotion hosting for my blog. May be i should also shift to amazon cloud.
S Anand 3 Jun 2013 7:30 am: I did try Linode early on, and it’s a fairly decent option. It’s just that for my needs, the $29/month that I can get with Amazon is too attractive.
Amit Chakradeo 3 Jun 2013 8:33 pm: If you don’t mind occasional downtimes and occasional ummm complete loss of data :-), you can consider low end VPS boxes. (http://lowendbox.com/) You can find some decent deals if you check out the posts there…
Sriram 15 Jan 2014 4:18 pm: have u tried openshift ?

Windows Live Local

Thu, 08 Dec 2005 12:00:00 +0000

Windows Live Local tries to match Google Maps. But Google Maps is just too fast. As I mentioned earlier (Why Google Reader): the reason I like Google is largely speed.>

500 mile e-mail

Wed, 07 Dec 2005 12:00:00 +0000

500 mile e-mail: the story of a server that would not send e-mail beyond 500 miles.