4 min read
In a recent article, the ChatGPT father and CEO of OpenAI, Sam Altman, claimed that “Please and Thank You to ChatGPT costs tens of millions of dollars”.
Altman’s suggestion of being impolite with machines to reduce costs might sound a bit unexpected. Still, Machine Learning algorithms require a considerable amount of CPU cycles, computational power and electricity to recognise the context in the natural language.
We should always avoid passing unnecessary information to machines if we want to optimise our requests, get responses quicker and waste as little electricity as possible.
Having a sustainable business is indeed a significant topic today, but it is sometimes viewed as another obligation (often a heavy one) to add to the list of regulations to comply with.
How can sustainability be turned into an opportunity?
What does it mean for a tech company to build a SaaS sustainable platform?
In this article, we aim to address these questions from a CTO perspective, which is not a common viewpoint to hear.
Let’s start with a simple poll: Do you remember this?
These are two SPARC ULTRA 1 workstations, developed by Sun Microsystems in 1995 before the acquisition by Oracle Corporation, and one X-TERM (just a terminal necessary to connect a monitor) on top of the stack.
At the time we are writing this article, it was 30 years ago!
The SPARC ULTRA 1 was one of the top workstations available on the market at the time, and one of the first based on the 64-bit architecture.
Each server has a CPU of 200 MHz, 200 MB of RAM (model ECC 60) and total storage capabilities of 50 GB on old-fashioned magnetic disks.
The native operating system is Solaris 2.5, but some Linux distributions are supported too, like Linux RedHat 4.2 SPARC.
To provide a comparison with modern devices, the cheapest smartphone available on the market is at least 5,000 times faster and offers 600 times the storage capacity.
So, why is the CTO of a modern organisation such as Spixii still using such old machines?
For sustainability!
Many programmers today just pass large datasets to libraries on Cloud infrastructure by simply "trusting" that the underlying software and hardware are performant and well-optimised.
On a SPARC ULTRA 1, optimisation is an absolute must.
The two (or more) units need to organise and distribute the computational workload in parallel: bottlenecks or a low level of parallelism will make any application extremely slow and unusable. Bad parallelism also results in high electricity consumption, as more CPU clocks and time are required to complete the tasks.
Large datasets are transmitted to the machines via a 10BASE Ethernet connection, which has a limited bandwidth of 100 Mbit/s. This is an excellent simulation of the real world, where fluctuations, partial network unavailability and general slowness happen frequently, despite the most advanced and stable connections provided by Cloud infrastructures.
Failing to broadcast the data through the network without chunking and recovery points implies resending the overall dataset, which is one of the most time-consuming activities of data analysis.
When it comes to data transmission, we often quote the former NASA Mission Control director Gene Kranz: “Failure is not an option”!
Finally, memory must be cleaned up and used carefully to avoid the exponential growth of indexes and hard-drive paging, which are the primary causes of delays in computation and, consequently, in overall time and electricity consumption.
As a result, if our code manages to run efficiently and quickly on such old devices, compile with high-level of paralleling paradigms, optimise networking exchanges and reduce access to the memory, then we can safely deploy it to a large cluster of servers on the Cloud and we are sure that it will work perfectly, will fully unleash the astonishing computational power of the Cloud infrastructure and will be very eco-friendly, with low-energy consumption.
This is how we ensure that our CPA platform is reliable, well-maintained and sustainable.
Conclusion
We gathered a lot of knowledge and awareness on sustainability by experimenting and simulating our SaaS platform on the SPARC ULTRA 1 units, where there is no room for mistakes or imperfect optimisations. Our learnings were summarised and incorporated into our ESG (Environmental, Social, and Governance) policies.
We are delighted that our clients have recognised Spixii as a trusted partner to improve their holistic sustainability approach.
