Reef Technologies

A fully remote Python software house

Productization of a financial analyzer

For a customer from USA, we optimized, extended and refactored an application performing analysis of financial situation of large companies. When we started working together, the product was a script reading and writing csv files, with the computation time on a large data set exceeding 35 hours. After we finished, all data was ingested and written to a database, with the calculations on the same data set taking about 40 minutes (on the same hardware).

 

Jay Newton, Director of Software Development, said:

Jay NewtonWorking with Reef Technologies was a great decision. We were concerned about the technical debt that we had and low-quality code, so we needed world-class professionals to make it up. We are extremely happy and satisfied with the results, since we managed to meet our really tight deadline. We will definitely keep up this partnership! Thanks!

Performance optimization

For a customer from Lyon (France), we investigated a server performance issue. Website seemed to be slow, despite a new dedicated server with 40 CPU cores, 256GB of RAM and a 500Mbps network.

The matter was urgent, a slow website was turning away paying customers, while an expensive marketing campaign was taking place. Money was being lost, we needed to move quickly.

The root cause was a large packet loss of SYN packets sent to a third-party vendor API. The most effective solution for this was to use persistent HTTPS connections, with many requests handled over the same socket. The change in the code was performed by software developers customer has already hired, based on our analysis.

Damien Gensanne said:

Damien GensanneI had a real complex and specific problem to fix (Network + Python + API problem), 4 programmers tried during one month and all failed! But Reef Technologies found it just in 2 days and explained exactly what is the problem and how to fix it. Real professionals and the best workers I’ve met in 2 years running my business!

consistent multi-client filesystem interface for low-cost cloud

A new low-cost could service (1/4th of AWS price) could potentially become an extremely cost effective storage asset for multiple bussinesses if it features an interface which is easy to integrate with their existing infrastructure.

Together with an open-source filesystem interface development team, we have designed many variants of a solution which would provide consistency and reliability in multiple endpoints using the same storage bucket.

The analysis has concluded that current feature setof the cloud in question does not allow for implementation of such a filesystem with acceptable performance characteristics. Using an additional service for locking would increase price too much and would potentially compromise security.

Additional features were requested from the could operator to enable implementation of a high-performance design.

X2 e-radio transmitter

X2 was designed as an advanced alternative to linux-hosted radio streaming programs available at the time.

The transmitter had full control over fade-in and fade-out, as well as on filtering the sound samples, which allowed for configuration flexibility so much desired by our radio operators. Thanks to this technology, they could separately configure the mixing of music-to-music and music-to-jingle, achieving unparalleled results in terms of end-user audio quality. X2 had also been one of the first transmitters to use a Compressor-Expander algorithm which carefully leveled the volume of music, delivering a smooth experience to the listener.

In 2005-2008 the X2 transmitter was licensed to the largest e-radio station in the world where it continuously delivered all input streams for 16 e-radio channels.

During all the years of production usage, X2 had no user-perceptible downtime, achieving 100% availability (including the period of partial power loss in the datacenter).

e-radio

We designed, implemented and deployed a custom e-radio system consisting of transmitter, relay network and a desktop player.

The transmitter used high quality audio files saved in lossless FLAC OGG format, while the end-user codec was a low-bandwidth AAC Plus v2. Despite (quite high) license fees usually required to stream and play the latter format, we were able to use an architecture which carried no license costs and was legal.

Additional cost reduction was possible due to network bandwidth fee optimization using dedicated servers from international datacenters.

Savings achieved this way summed up to $50.000 per month, making the entire project viable. This is how the first e-radio of such size in the world was born.

network trace dump analyzer

Serving large files to a nation-scale audience is a challenge. It is even harder if the downloaded file is a binary installer and a considerable percentage of users use an old version of Internet Explorer which reports a file only partially completed as finished successfully.

Our customer had an issue with shipping their binary installer to his users.

In 2009 we have designed and deployed a raw network dump analyzer in order to diagnose issues with the file-serving infrastructure. Thanks to this, several issues have been identified and fixed, reducing the failed downloads from up to 80% down to a few percent (almost exclusively gprs users and user-cancelled transfers were left).

CPU checker for flash-based advertisements

Websites and desktop application vendors that sell flash-based advertisements often have issues with performance by the ads. This happens because the ad buyer uses vendors to prepare the content that looks stunning. Testing for CPU consumption is not a standard practice. If the performance of application is poor (due to a CPU-hungry ad), the reputation of the website/application suffers.

In 2010 we designed, reviewed and deployed a system which used a common desktop computer along with a task queue, API, web panel and email notification system. This system was used extensively by the Ad Traffic team of our customer to verify CPU consumption of the content before it was accepted for emission. The system operated continuously for many years.

In result, performance complaints for the affected services have vanished and the reputation of the customer company was gradually recovered.