[{"content":"In 2023 I moved this site to AWS. As part of the move, I set up request logging, which means I can see every single request ever made to the site. Three years on, it feels like a good time to look back over the data.\nThe site is effectively dead. I haven’t published any new content in that time, so any traffic it gets isn’t coming from people following a new post or a fresh link. That makes it a useful little laboratory: whatever shows up in the logs is background noise of the internet, with nothing to drown it out.\nHow the data is stored\rThe logs are written as compressed files into an Amazon S3 bucket. From there I pull them into Amazon Athena, which lets me run SQL queries over the full data set without having to download or unpack anything. Every question I had about the site turned into a query. Scaning the full data set tasks about 11 seconds to proses the full 200MB of logs.\nRequests per month\rTo get an overview, I started with the simplest question: how many requests did the site receive each month?\nWhat stands out is the growth. Requests have climbed over the three years, even though there has been no new content. That means essentially 100% of that growth is automated. Nobody is discovering the site and telling their friends. The bots are just getting busier.\nThe top 25 URLs\rNext I looked at the 25 most requested URLs.\nThis list shows clear signs of automated scanning. Two entries in particular caught my eye:\n/wp-login.php a check to see whether the site is a WordPress install. It isn’t, but the scanners ask anyway, because on the internet at large it is a very good bet. /.env a file that often holds API keys, database passwords and other secrets. If a site accidentally serves it, the attacker gets the keys to everything. I’m curious how often that request actually succeeds on the wider internet. User agents\rThen I looked at what the visitors claimed to be, using the user agent string sent with each request.\nThe headline number: over 30% of all requests had no user agent at all. A real browser always identifies itself, so a missing user agent is a strong sign of a script.\nAmong the ones that did identify themselves, the tooling is revealing:\ncurl/8.7.1 Go-http-client/1.1 Go-http-client/2.0 python-requests/2.31.0 These are all command-line tools and programming libraries, the sort of thing you use to write a scanner in an afternoon. The remaining user agents looked like ordinary browsers. That doesn’t mean they were real people, since faking a browser string is trivial, but they were at least plausible.\nWhere is the traffic coming from?\rTo get a rough idea of geography, I grouped requests by the edge location (PoP) that served them. This isn’t a perfect measure of where a request originated, but it’s a decent proxy for where in the world the traffic entered the network.\nCDG52-P4 is Charles de Gaulle International Airport in France YTO53-P1 is the Greater Toronto Area in Canada AMS58-P5 is Amsterdam Airport Schiphol in Netherlands Beyond those, the traffic is spread across a wide range of locations around the world.\nIt’s also worth noting where the traffic isn’t. If real people were looking at this site, I’d expect to see the locations closest to its audience near the top. Those locations sit well down the list:\nRank PoP Location Share of traffic 11 SYD3-P3 Sydney 2.7% 19 SYD62-P3 Sydney 1.5% 130 AKL53-P1 Auckland 0.04% 132 AKL50-C1 Auckland 0.04% Auckland barely registers. The traffic follows where the data centres and scanning infrastructure are, not where the human audience is.\nThe most persistent visitor\rThe single busiest IP address was in France, responsible for a little over 0.7% of all requests. That sounds small, but for one address among everything the internet threw at this site, it’s a lot.\nFirst seen: 2026-04-12 Last seen: 2026-09-16\nIn that time it completed four full scans of the site, working methodically through a long list of paths. Some of what it asked for:\n/etc/passwd /id_rsa /.git/hooks /opt/mailcow-dockerized/mailcow.conf /sendgrid.env That list is a snapshot of what attackers hope to find lying around: system password files, private SSH keys, source control data, and configuration files for mail servers and email services. This is credential fishing, plain and simple. The scanner isn’t targeting me in particular. It is casting a wide net across the internet and hoping some server is misconfigured enough to hand something over.\nWhat this tells me\rOverall, this is a large data set of almost purely automated activity. A site with no content, no audience and no promotion still attracts a steadily growing stream of scanners, scrapers and credential fishers, and the volume is rising year on year.\nIf you run a website of any kind, this is a good reminder that someone is already checking it for leftover secrets, whether or not you’ve ever told anyone it exists.\n","date":"2026-10-25T00:00:00Z","permalink":"/p/3-years-of-bot-traffic-analyzed/","title":"3 years of bot traffic analyzed"},{"content":"I have moved my website hosting to AWS. This was done for two reasons. Improve load times and lower running costs.\nThe old stack: Vultr + Epic + OpenBSD + httpd\nIn the server, I was running OpenBSD. I started off on this as I wanted to self-host an email server. In the OpenBSD VPS, I was using httpd, openSMTPD, and dovecot. This stack made hosting an email server not as painful as using Postfix. As with all email servers I had lots of delivery problems. In the end, I moved to Protonmail for my email hosting, letting me keep my custom domain without the maintenance and delivery problems.\nThe VPS was in Sydney. This gave Oceania good response times. The single point of presence did limit response times in NA and EU. As I was on the cheapest VPS with an ipv4 address I was paying $5 per month. Working out to $60 per year. This is quite a lot for a website that gets less than 10 views a month.\nThe new stack: AWS Route53 + AWS S3 + AWS Cloud Front\nI have now moved to AWS. To do this I am using 3 AWS services. Route53 for DNS. S3 for storing the website. And Cloud Front as a CDN. Cloud Front does all the SSL and cacheing.\nThis set of tools gives worldwide CDN coverage, less downtime, and costs less.\nAs my traffic is so low, I am in the free tier for most of the services that I use. Costs have settled to $0.50 per month. This makes the total cost $6 per year. That is almost what I paid a month with the old stack.\nOnce I moved from Epic to AWS Route53 for my DNS I had problems linking my domain’s apex to Cloud Front. It would always give pops that were in the EU. After months of research, I found that I had not changed the name server to the ones my hosted zone used. After I made this change the system is now working with 100% efficiency.\n","date":"2023-09-20T00:00:00Z","permalink":"/p/website-move-to-aws/","title":"Website Move to AWS"},{"content":"I have now completed my Bachelor of Applied Information Technology. It took me three years to do all the course content. Over this time, I have made many lifelong friends. I had an average grade of A+. My favorite paper was COMP605, Data Structures and Algorithms. I found the content clicked with me.\nOver this time, I also took part in the Applied Deep Learning Research Group. As part of this, I helped with the paper Automatic counting of chickens in confined area using the LCFCN algorithm. I also joined the Peer Tutor program. Near the start of my time as a tutor, there was a third-year student in need of tutoring. As they did not have any third-year tutors, I was asked to step in and tutor him despite the fact I was a year behind him. He needed help with programming in the language php. In the course we had not yet covered php but, I had experience from outside of Wintec. I was able to apply my knowledge to successfully help him pass the paper.\n","date":"2022-07-03T00:00:00Z","permalink":"/p/bachelor-of-applied-information-technology/","title":"Bachelor of Applied Information Technology"},{"content":" Over the last year I have been working on a research paper with some of my tutors at Wintec. For this paper we did research into LCFCN. You can read the paper here.\nI started off with an implementation called Deepfish. I added support for Microsoft Windows. Documented and built a consistent installation script to enable reproducible installs. I also implemented a custom loss function for the network. I developed wrappers so that my custom labeling tool could generate predictions and make training data correctly.\nI developed a custom point level annotation tool in Python. This enables the user to quickly label all the objects in an image and save it to an extensible text-based format. The tool used OpenCV to display the images. It was put to the test labeling more than 1000 pictures. Due to the annotation format being lightweight, sharing annotations between collaborators was easy.\nOver the research we used a variety of open-source resources and on-site photography to make custom data sets. To extend the data sets data augmentation techniques were used. Due to the architecture of the network, it cannot distinguish between an image and a mirrored copy. This enabled us to 8X the data we labeled.\nOne of the data sets that we used we have published. It is a data set of cows labeled from aerial photography provided by the New Zealand government. All the pictures are 500 pixels square and have a density of 10 centimeters per pixel. The data set is hosted here.\nI volunteered my personal computer to provide the compute power necessary. It took multiple days per dataset to train the network. Once trained the network would then be tested against a final holdout validation set. I wrote some scripts to take the final numbers and compute overall loss values. Participation certificate for ISCV2022\nThe paper was then presented at an international conference. I received a certificate of participation for my contributions to the paper.\n","date":"2022-07-01T00:00:00Z","image":"/p/lcfcn-paper/splash.jpg","permalink":"/p/lcfcn-paper/","title":"LCFCN Paper"},{"content":"I have joined the Applied Deep Learning Research group. We are working on revolutionizing animal welfare. Our project combines satellite imagery with artificial intelligence to create a cutting edge automated dairy management system.\n","date":"2020-10-02T00:00:00Z","permalink":"/p/deep-learning/","title":"Deep Learning"},{"content":" I made this game in Games Development (COMP706). It is a 2D platformer. I made the movement code for the main character from scratch. The tiles in the levels were made using a combination of 3D and 2D graphics. I used Godot for the game engine as it has a good mix of community support and ease of use. I ended up with a polished final product and got A++ for the game.\nYou can play the game for free right now here.\n","date":"2020-04-21T00:00:00Z","image":"/p/danger-dag/splash.png","permalink":"/p/danger-dag/","title":"Danger Dag"},{"content":"Mistype is a type racer with a twist! Many of the words are spelled wrong. You must fix the typos at speed. You can race a bot or a friend. Ready, set, type!\nI made mistype with my friend Baxter in the Global Game Jam 2020. We set up and used a git repository to manage our work over the 48 hours. We were the first group to make a playable prototype at the Waikato University venue.\nIt was playable online but link rot takes us all\n","date":"2020-02-08T00:00:00Z","image":"/p/mistype/splash.png","permalink":"/p/mistype/","title":"Mistype"},{"content":"I started using Blender in 2015. Since then I have spent many years messing around and have moved from not being able to click on objects to making complex 3D models. In fact I made the banner at the top of this site. Here is another render I made\nIn 2017 I was asked to make a short animation for One Hill of a Fight, a play that explains the development of the Hamilton CBD. In the 1930’s the Borough Engineer constructed a mechanical model to convince the Council to remove a central hill and develop Garden Place. My animation recreated that model based on a description of the original.\nYour browser doesn't support HTML5 video.\rHere is a link to the video instead.\r","date":"2017-08-17T00:00:00Z","image":"/p/3d-renders/ocean.png","permalink":"/p/3d-renders/","title":"3D Renders"},{"content":"I made a version of Camel Up to learn about pointers in C. Camel Up is a board game where camels race. When a camel moves it will move and bring along all the camels on top of it. If a camel moves to a square with a camel then it will get on top.\nThis taught me how to reference and de-reference pointers. I made a struct to keep a camel. Each camel struct hold its name and which camel it is riding. This means that if a camel below moves then it will too.\nThe source code is here.\n","date":"2016-12-25T00:00:00Z","permalink":"/p/camel-up/","title":"Camel Up"}]