Supercomputer Rack Development

Rack development started in unusual way: I came for an interview in T-Platforms and I was sure that I would work as a PM under some kind of blade server. However, when I mentioned that in MSU I studied 12 additional supercomputer courses, it was some kind of a surprise for employer. Then, I was offered a PM position, to start RnD project with 256 nodes rack, 300 TFlops performance. I exclaimed "But it will be most powerful rack in the world!" and got calm answer "Yes, so what?"

 I wouldn’t say that my work started in standard way. I was provided with a table and a notebook. That was all I had at the start. No, not only this. I also had a right for 4 hours a week for Lead Architect consulting time. I asked "Sorry, but where is my team?" and got answer "You have no restriction, guy. You can hire anyone you want from outside world. Our engineers are overloaded now, sorry".

I found this challenging in true sense of this word. There were no restrictions exclude ones in my mind. At that time, I already had an experience of acquiring a team from the scratch. Generally, I started to delve deeper into the rack concept. It took about couples of weeks - I known hardware world for consumer electronics, but server world is different.

Usual practice I had at that time: I could ask any question to well-experienced engineers, and had instant answer. The only restriction was that communication was limited by one minute. Therefore, maximum what I received was required internet link or answer like "yes-no". No any explanations, just answer. Funny time!

Well, that’s enough for pre-history, let’s consider the rack itself. This is a final rack view:

rack new


Let's go to the descriptions about rack components (follow the links below):

Rack Sectioning - how the rack is constructed, to have as liitle SKU as possible

Rack Control - how the rack's (and other racks) components are managed

Rack Networking - network description and structure, inside and outside the rack

Rack Powering - how the power for rack components creates and distributes between nodes

Rack Cooling - how we remove heat from the rack and what we do next with this heat

In addition, separate - about Server Motherboard, how its components interact to each other

There is no NDA disclose, all pictures were taken from official presentation on T-Platforms web site

©2021 Nikolay Bodunov. All Rights Reserved.