Over the past ten years, we've faced our fair share of challenges. We believe in being open about them — because it's through those experiences that we've grown, evolved, and continued to innovate.
1 – Not all CPUs are equal.
Each CPU of a specific SKU has its own unique characteristics due to the manufacturing process, making them all slightly unique. Certain CPUs may need more voltage than others to achieve the same frequency or may not even be capable of maintaining said frequency.
2 – Shift from low to high core counts.
Over the years, as market data feeds have continued to explode in message rate and the number of data points of interest to traders has increased, higher core counts have become more desirable, even at the cost of lower frequency or performance. Having the right data and the right strategy has become more important than simple, outright speed.
3 – Packaging & Logistics are important.
We’ve all seen how parcel couriers handle our online purchases, now imagine them doing that with a 25 kg (55 lbs) high-performance server. Whilst we made significant changes to our packaging to withstand as much punishment as possible, ultimately, we decided to swap all shipments to pallet-based services to ensure that systems were delivered safely and handled with care. Whilst more expensive, it’s better than receiving a damaged server.
4 – Newer architecture isn’t always better.
We all want the latest and greatest, but sometimes new technologies don’t always align with the same goals as Blackcore and our clients' requirements– going faster! The most recent example of this is the Intel Ultra 9 series of CPUs, which moved to a more power efficient design but sacrificed some raw latency performance.
5 – Size isn’t everything (1U vs 2U).
Historically, co-location racks were charged by the number of bays (or “U”) consumed. Or, if you owned the racks, then it was prudent to fit as many systems per rack as possible to reduce the overall rack footprint. This means 1U systems were preferred. However, as system power draw has increased, and data-centre power became constrained, a move to charging for power used rather than space emerged. We decided that 1U systems had too many drawbacks – reduced performance, increased heat, reduced reliability – and the benefit was no longer there. As such, we drove an engineering-first approach to move the overclocked server market to 2U and up.
6 – You can’t do too much testing.
Testing goes beyond “does it boot?”
At Blackcore, we have always been transparent with our testing processes. We’ve evolved these tests over the years, and now every system must pass our stress tests for a minimum of 16 hours – twice! We know this is not an industry standard practice, and whilst this requires more work and takes longer, we have seen significantly lower dead-on-arrival rates and improved client experience.
7 – Prepare for failure, engineer to avoid.
No matter how well you test, some percentage of hardware will fail. We've learned that how you deal with these situations can massively impact the client. We designed our warranty packages to not waste time – such as needlessly sending engineers to site – and provide clients with the tools to have their own agency in times of failure, by offering a spares program for clients to hold cold systems ready for use at either our own facilities or theirs.
8 – Enterprise tooling is critical for success.
Clients need to manage their deployed systems efficiently, and largely this means being able to manage them pragmatically. We learned that web-UIs and manual processes were not sustainable workflows and so began investing in having functional IPMI and Redfish capabilities as well as ensuring we could provide Linux-based tooling to interface with our motherboards. We also took the steps to ensure we have the proper licensing in place to provide our clients official tooling from vendors like AMI to interface with the BIOS.
9 – CPU frequency isn’t the only important metric.
When we first started, CPU frequency was everything – the faster the better, or so we thought. However, unless you look at multiple system metrics, how can you understand the effect that increasing CPU frequency is having? Additionally, is CPU frequency the most important metric? As we now know, CPU cache performance is a significant metric for electronic trading algorithms. Maximising all relevant metrics is one strategy, or there may be situations where sacrificing something like peak frequency may provide benefits in other metrics. Regardless, Blackcore likes to be transparent, and so we publish benchmarks for PCIe, cache, and RAM latency on all our spec sheets.
10 – Not all users have the same requirements.
“One size doesn’t fit all”
A fundamental part of the Blackcore ethos is flexibility, with a focus on providing the right solution for the client. Some clients want extensive pre-configuration done prior to shipment, such as IP configuration, specific labelling, or asset tracking documents uploaded to a website. Whilst others want something barebones that they can implement into their existing automated deployment infrastructure.
We can sum up these lessons as:
-
Partner with our clients
-
Take their feedback and harness it
-
Continuously improve processes and procedures
-
Be easy to work with
All of which has helped us deliver the best product on the market.
If you would like to learn more about Blackcore, please email [email protected] or book a call with our CRO, Ciaran Kennedy here.
James Lupton