Scaleway disarms its ARM64 cloud, cites unreliable hardware as the reason

One of the few clouds to offer 64-bit Arm-compatible servers is dropping the architecture. Scaleway revealed new Arm-powered efforts in June 2017, offering four instance types running on Cavium, now Marvell, ThunderX SoCs. At the time of launch, Scaleway said the launch was a culmination of its long efforts to make Arm- …

[Bronek Kozicki]

1st gen Cavium TunderX to retire

It's an old chip and I am not surprised that it is being retired. Perhaps they waited for a worthy successor - there are few on the horizon as of now, but it will be little longer before the new hardware can be installed in the datacentres. They also won't be cheap, which makes them hard sell for bare-metal. In the meantime, the old bare-metal instances are turning out to be pain for the users and maintenance staff which is not surprising.

[steelpillow]

Hazard a guess...

...that the Universal Reason is at least partly to blame - the entry of new, more reliable servers to the marketplace has been delayed beyond the end-of-life of the current ones.

[Peter2]

Re: Hazard a guess...

It's probably a combination of expensive replacement, few users and having to maintain the skills to support it costing more than the service is worth to the business.

With that in mind, if there are other people offering this then Scaleway withdrawing from the market will help consolidate the user base to another provider making their offering more economically viable.

[Phil Endecott]

I tried to use this for a few months and it was indeed horribly unreliable.

It wasn’t unpopular. One of the problems was that for some configuration changes you needed to snapshot, shutdown, reconfigure and restart from the snapshot - and at the restart step, you could discover that there were no more instances available and you’d be in a queue - perhaps days long - for someone else to release one. Not ideal for a business. I learned a lot about how to make resilient systems, mainly by failing over to AWS when the Scaleway server crashed.

The AWS ARM instances are good and demonstrate what is possible; but maybe it’s only possible if you can make your own chips, as Amazon has. Maybe the off-the-shelf servers are not good enough yet. (My own experience with a Gigabyte ARM server motherboard was as poor as the Scaleway experience).

Scaleway’s earlier 32-bit ARM hardware was apparently reliable, though low-spec.

[The Man Who Fell To Earth]

Yet another example

Of "cloud computing" being untrustworthy. Doesn't matter if its "reliable" if you can't trust that it will "be there".

[bombastic bob]

Re: Yet another example

well cloudy servers have their place BUT yeah, "all eggs in one basket" shouldn't be done either. Somewhere a balance of that would make the most sense. Maybe the cloud system includes a server local to the company on its private connection, and then a good portion of the real work is done "in the cloud" on other servers... so if THEY go down [or YOU], the service is not significantly interrupted - ok performance degraded, but not interrupted. That's the point. Just needs to be properly designed for failure tolerance and failover.

[Stuart Halliday]

Bad Motherboard design or dodgy chip sets?

[Scene it all]

I'm thinking the boards. Poor quality control in soldering, etc? Poorly thought out cooling?

[bombastic bob]

stuff that is "too cutting edge" sometimes fails earlier than expected. I've had to replace a couple o' motherboards for similar reasons... and the 1 year later motherboard is rock solid, performs the same, and lasts for over a decade [and costs less]. Could just be THAT.

[bombastic bob]

If 'power sipping' ARM64 costs less... won't people want it?

I think it may be that they did not properly investigate the marketing opportunity, to use ARM64 to provide CUT RATE services.

Or maybe they did, and rejected the idea and didn't tell anyone?

in my opinion it could justify the procurement of lower cost ARM64-based cloud servers, which should run cooler, use less 'trons, and maybe even COST LESS for the hardware itself, as compared to its amd64 architecture cousins. Just a thought, anyway. Maybe some other cloudy provider can step in and do this, and take all of the business.

So - what's the highest expense for a cloudy provider:

a) electricity

b) intarwebs bandwidth

c) hardware

d) administrative (building, people, legal, gummint, ...)

I think a and c can be mitigated with arm64. 'b' can't though. So knowing if 'b' is the limiting factor might explain things. [of course 'd' is probably a fixed cost, and I'd expect IT support to be part of that].

[AConcernedCitizen]

Re: If 'power sipping' ARM64 costs less... won't people want it?

I personally think that a lot of customers reject ARM64 regardless of price from the simple fact that companies (like Microsoft) generally only offer x86(-64)-compatible binaries of their products.