Professional Portfolio Web Applications and Software I made for others

Summary

Penny auctions are a high intensity type of auction which are always very close to ending, with a timer that keeps extending itself a few seconds every time someone places a bid, while dozens of users in each auction try to outbid each other at the last possible second. Because of this, penny auction sites have very strict performance and latency requirements. Also, all clients need be updated constantly, within less than a second, or they might miss the opportunity to bid on time. And like many of these projects, it happened before WebSockets were widely supported.

When I was contracted to work on this project, there was already a basic live site, which wasn't working too well and was starting to have performance problems due to the growing traffic. The code at the time wasn't well structured and didn't lend itself to easy modifications, or to fast performance. Over time, piece by piece, that code was almost completely rewritten.

In the process I completely rewrote the polling mechanism on both client and server, leading to much faster performance, an enormous reduction in bandwidth, and a cooler server. In the end we were able to expand the maximum number of users connected simultaneously by 8x, while reducing server costs by 54%.

Besides the performance improvements, over time I added several new features to the site, most notably an "experience" system, to give rewards to high profile users of the site, automatic refunds for unhappy customers, and "multi-item" auctions, which allowed people to pick between several items after they won.

Highlights

• Performance: The site runs in a single, pretty standard, virtual server with 2010 hardware. This server runs nginx, MySQL and memcached, and supports about 500 users at the same time, all polling twice a second, with CPU around 10%. There is plenty of margin for growth before this code needs to be revised.
• Memcached: The site relies heavily on memcached to be able to react fast. The amount of memory that memcached is using is ridiculously low, however each item gets read an enormous number of times every second, making memcached crucial to performance. Absolutely all polling requests are processed without touching the database.
• Bandwidth: The polling is extremely efficient in terms of bandwidth. Users poll with only a specific ID assigned only to their window, and the server replies with only the information that has changed, making the typical response body about 70 bytes, in a homepage with 50 auctions that are constantly getting bid on.
• Transactions Handling: The codebase has successfully managed over 20 million dollars in transactions and keeps account balances for hundreds of thousands of users, dealing with several payment processors, automatic refunds, and hundreds of thousands of internal balance operations, bids, bets and payouts every day.