Upgrade: join us inside the datacentre

On Tuesday 3 May 2022, we upgraded our infrastructure. Details and photos: we are taking you with us, right into the heart of the datacentre!

A new datacentre extension

Our Cyrès datacentre (in French) in Tours recently expanded with a new 250 m² room. This coincided with our own growth and new requirements.

To keep growing, we therefore chose to move our servers into this innovative new extension, built around hot aisle containment (in French). In practice, the racks are almost airtight, and a constant airflow is directed from the cold side (the front of the racks) to the hot side (the back of the racks). When you open the door to step into the hot aisle, you can feel a remarkable draught. The result is far better cooling of the machines and top-level energy efficiency.

Ultimately, this technology developed by EFIRACK makes for more energy-efficient operation. Find out more in this video (in French).

Technical choices and preparation

We took the opportunity to replace our network equipment in anticipation of future developments.

Over several months, our technical team selected, ordered, tested and configured each new piece of network equipment. We also designed and tested several ways of organising the power and network layout of the racks, and planned every cable, even every cable tie, in advance so that we would be ready on the day.

A major strength of HaiSoft is the versatility and cooperation of its team members: every technician has networking skills in addition to their Linux sysadmin speciality. The whole team therefore takes part in the discussion and agrees on the equipment that it will install and use itself afterwards. I think I can speak for everyone when I say that this situation, which is only a dream in many other companies, makes the whole HaiSoft team very happy.

For our switches, we replaced Cisco with Mikrotik, whose equipment had fully satisfied us during several years of testing. The brand now offers some of the best-equipped switches on the market. Ours feature: built-in dual power supplies, 48 Ethernet ports, 4 SFP+ ports (10 Gbit/s) and 2 QSFP+ ports (40 Gbit/s). The switching capacity is 168 Gbit/s per switch. If that means little to you, simply put: you cannot hope for better performance given the available ports. Everything can run at 100% simultaneously without any loss of throughput.

We interconnected the switches at 40 Gbit/s via QSFP+. This rules out any risk of saturating the internal network and makes 10 Gbit/s links between certain servers possible.

Combined with the newly available server slots, this takes us one step closer to new types of hosting-related services. We look forward to unveiling them to you soon.

The preparatory work

Physically migrating machines requires a great deal of organisation and close attention to detail.

On 2 May 2022, part of our team therefore went to the datacentre to prepare: installing and interconnecting the network equipment (switches), carrying out final checks, and finally pre-cabling network and power connections ready for the servers.

The final operation

Tuesday 3 May from 8 pm was the time scheduled, and announced by e-mail, for this exceptional service interruption.

An operation requiring downtime is always an exceptional moment in the hosting world. Working in the evening keeps the inconvenience to a minimum, but we nonetheless optimised the migration process as much as possible to reduce the length of the interruption.

To do so, we chose to shut down the servers in small groups, physically move the machines into the new racks, plug them in and power them back on, and check that all services were accessible before moving on to the next group. We had announced 30 minutes to 1 hour of downtime. In practice, we worked efficiently and everything went without a hitch: the average interruption was around 15 to 20 minutes.

This was only possible thanks to our close-knit, well-coordinated, energetic and efficient team:

• Damien: supervision, shutdown requests, unracking, monitoring
• Benoît (remotely): sending shutdowns, monitoring, post-reboot fixes
• Jérôme: physical moves, reracking, power-on checks
• Robin (author of this article): reracking, connecting the servers, organising the racks

I would like to take this opportunity to thank the team for making this smoothly run operation possible; I will remember it for the rest of my life.

At around 12:30 am, after more than 4 hours of work and tidying up, the team, though a little tired, was all smiles, and with good reason: everything was working perfectly, exactly as planned.

We hope you enjoyed sharing this experience with us.

Remember to follow us, and feel free to share your feedback on social media.