The first customer instance booted in Amsterdam on 19 August. Frankfurt followed a fortnight later, London at the end of September, Bucharest last week. Four sites in eight weeks, which was either ambitious or reckless and we are still not certain which.
This is the first entry in what is meant to become an engineering diary. Whether it survives contact with actually running the business is an open question.
Why these four cities
Amsterdam because the peering is dense enough that a small operator can buy good connectivity without buying scale. Everything we build gets built here first and everything we test gets tested here first, and that is unlikely to change.
Frankfurt for the single hop into central and eastern Europe, which is where a surprising share of early interest came from.
London for the transatlantic path. It is the shortest hop to North America in our European footprint and it will stay useful even after we have a site on the other side.
Bucharest for the price per terabyte, which is better than anywhere else we looked by a margin that made the decision easy. Cheap bandwidth and generous transit, without the compromises we expected to find attached to it.
What we built
Dedicated cores from the first day. Not shared, not bursting, not a fraction of a core with an optimistic name. A core you buy is a core nobody else is scheduled on, and there is no oversubscription ratio to disclose because the ratio is one.
The reason is commercial rather than principled. Overselling is a bet that your customers will not all be busy at once, and when the bet loses, the thing that fails is the product every one of them bought. We would rather sell fewer cores at a price that works.
Storage is NVMe in mirrored pairs. Networking is ten gigabits per port. Sign-up takes an email address and nothing else, which is the part people write in about most.
The bad decision
We launched accepting Bitcoin and nothing else, on the reasoning that it is the asset everyone has.
The first ticket asking for Monero arrived twelve days later. Another followed the same afternoon. Within a month it was the most requested thing on the list, ahead of any feature, and by then we had also watched a customer pay a six euro network fee on a twenty-nine euro invoice, which is not a product, it is an apology with a server attached.
Monero and Litecoin are being added now. Several more assets follow, and the lesson we are taking is broader than the specific coins: our opinion about which asset is sensible is worth less than the customer’s, since they are the one paying the fee.
The other thing we will probably regret
We shipped a four-euro plan, because everybody ships a four-euro plan.
It is already the source of most of the support we do, which we assumed would be true and did not expect quite so quickly. Whether it earns its place is a question for a later post, once we have twelve months of numbers rather than eight weeks of impressions.
What we are committing to now
No identity documents, at any order value, for any reason. An email address is the whole of what sign-up requires and there is no threshold at which that changes.
Refunds without an argument, in the asset you paid with. A price list that does not depend on who is asking. And a journal, which is this, kept honestly enough that the entry about the four-euro plan will get written even if it says we were wrong.
Next site is Helsinki, for reasons that will be obvious to anyone who has seen an electricity bill.