xnbd-client --blocksize 4096 --retry=900 10.1.32.197 4096 /dev/nbd0

  root@scw-6d2eff:~# df -h
  Filesystem      Size  Used Avail Use% Mounted on
  /dev/nbd0        46G  507M   44G   2% /

Like EC2 and others, the box has an internal ip allocated and an external ip via nat (or something else).

I was able to download a file from OVH with ~500 Mbit/s throughput when writing to the filesystem. Writing to /dev/null it increases to ~650 Mbit/s.

Running an untuned nginx, it can do ~300 req/s with ~20% cpu util.

Seeing as the storage is still virtual, I'm not really sure why this would be so much better than a virtual machine. I like the idea though.

I used to work on Rackspace OnMetal: http://www.rackspace.com/cloud/servers/onmetal

https://journal.paul.querna.org/articles/2014/07/02/putting-...

In OnMetal this not the case -- you get local storage, you have your local kernel that you can re-install or do whatever you want to -- and if we have a control plane outage, you can reboot -- there is no dependency once your server is created on a provider hosted PXE server.

OnMetal isn't cheapo-ARM servers, but it is possible to build a "bare metal" cloud, which is truly bare metal, like you would normally get, and no remote block stores for your root FS.

OpenStack Ironic[1] can do this today, given you had right hardware. Making that hardware "mutli-tenant" is not a trivial project, mostly because the firmware stacks are still very closed, but it can be done. Top of Rack switches are also another area of complexity/pain, but also feasible to overcome with a little engineering.

[1]: OpenStack has its complexities, but the OnMetal project has upstreamed everything feasible we can.

https://www.packet.net/blog/how-we-failed-at-openstack

It'll be doing better than a typical virtual machine because there's no contention on the host side, kernel or IO, and at this point, you've probably got the storage device to yourself so there's no bus contention.

The trade-off is shutdown time: they've got to slurp your disc image off to their block store before they can release your node back to the pool.

> Each server has access to a pool of local drives. These drives are exported to servers via the NBD protocol which effectively makes them network drives. However, network between these drives and C1 nodes are dedicated PCB tracks which ensures minimal latency and avoids network congestion.

> There is no redundancy on these volumes, you need to handle redundancy on your side! They are archieved to a permanent storage when you start and stop your server.

> Local volumes are 100% SSD drives which are able to deliver a lot of IOPS and are perfect for random read/write patterns. The maximum size of LSSD volumes is 150GB.

Found on https://www.scaleway.com/faq/server#local_volumes

I wouldn't trust the packages for ARM because there are variations. Even GoLang doesn't work on all ARM processors, only some!

True dedicated servers would be much more powerful, and frankly, this is a service form Online.NET.

http://www.online.net/en/dedicated-server/dedibox-classic

For 3x the price, you can get a real Xeon processor [that will likely perform equivalent or better to 3x scaleway nodes] and 8x the RAM. At this price point, I don't see the appeal outside of the object storage.

They're very reliable, and a pretty good competitor to KimSufi/SoYouStart/OVH.

Iliad, the group behind Online.net and Scaleway, also owns Free Mobile and Free Telecom and Free Internet, all of which are quite popular.

All the network properties operated under the Iliad umbrella suffer from this. Iliad and Orange pushed a bill through parliament that basically says "French ISPs can charge you whatever they want and you're required to pay it because operating in France means mandatory connections to the entrenched regional ISPs"

we are also adding a new IX soon (order was sent yesterday) which will improve our peering towards Europe

the infos you have are at least 10 years old imho... maybe look at http://map.online.net

You may want to make that map more obvious/prevalent on Online.NET, I wasn't aware you had a public network map like OVH does.

Bold clame I see. We begun offering the exact same configuration early last year ( http://www.unixy.net/arm-server/ )

But, you know, salestext is useless :)

"Average delivery time" 90 days? I'm not even sure what that means, but it doesn't strike me as particularly useful...

The only thing I could not find quickly is info about the company: Is it a startup? Is it funded?

Until I found out it's a new product by Online, which is a subsidiary to one of France biggest telecom disruptor. There should be an About us that tells just that, which would clearly give more confidence to potential customers.

Except when you add a credit card, they try to bill you for 20EUR right away.

This is a fairly common practice to weed out stolen credit-cards and to make sure you are real subscriber/customer, no biggie.

I would recommend changing that behaviour ASAP ;)

I use MariaDB-Galera in a project of mine to be able to run it in a parallel fashion. To determine if the servers were suitable for my use, I tried to install MariaDB-Galera, only to find out:

- MariaDB doesn't officially support ARM yet;

- The Debian and Ubuntu repos don't have the MariaDB-Galera edition available;

- All repos I could find with the Galera edition didn't have it compiled for armhf, including the official repo, by virtue of the first point.

- Compiling it on the server resulted in other problems: besides the usual compile errors that can be more or less sorted out by hand, and the dependency hell of the resulting deb packages (remember, no armhf packages available) I figured it would be unacceptable (in terms of ease-of-maintenance) to have to recompile MariaDB every time a new version with security fixes came out. This is something I can live with when toying with Raspberry Pis and similar things, but not on production stuff.

I eventually gave up, declaring the servers unsuitable for my use. End of anecdote.

IMHO they are charging too much for the product, at a time when the ARM ecosystem is not yet ready for server use (the architecture and OS support are fine, the user-land software, not so much). The servers are bare metal (it's awesome being able to use as much CPU as you want without breaching AUPs or hitting artificial caps), but the networked storage, while definitely better than HDDs, may not fulfill the needs of those with higher I/O requirements (I admit I haven't tested, also by virtue of Galera not being available). Lastly, the prices first show without VAT, which means that if you're in the EU and without a VAT ID, it'll turn out to be even more expensive. I'd consider something like €7/month (perhaps with less storage) to be a more acceptable price.

Fwiw it's €10/mo if you leave it on all the time, which isn't too much higher than that. The headline pricing of €0.02/hr is capped at a max €10/mo, i.e. free past 500 hours. Sort of similar to how Google Cloud does automatic discounts for high-utilization instances, except that Google's utilization discounts are a curve rather than a cliff. (But yes, still not including VAT.)

[1] https://packages.debian.org/search?arch=armhf&keywords=maria...

A couple of friends of mine are happier with compatibility than I am, too. Others are even more disappointed than I am, since one of the uses they have for dedicated stuff is game servers, which usually are closed source and not available for ARM.

In practice, you're stuck dealing with the sort of headaches that come with doing something unusual, like repos not having packages you want. Overall it seemed to work reasonably during the beta but I'm surprised they weren't more aggressive with pricing.

Hard to argue with the value if you are actually using the bandwidth, but otherwise not that interesting an offer from a practical perspective.

Kimsufi offers true dedicated Atom boxes with SSDs for 10 GBPs / month.

I also think that if you're "using the bandwidth", a $5/month Digital Ocean VM is probably a better value.

VMs are also bad at some specific tasks - they have issues with clock accuracy for example.

One appeal of ARM, specifically, would be that it's cheap. Sadly these are based on ARMv7, which is a lot more limited than aarch64/ARMv8 and Intel CPUs. - but hopefully they or someone else will have an ARMv8 offering out soon.

You'd be better off getting one of Online.NET's 30 euro nodes and using the Object Store from Scaleway. That is the only part of this announcement that appeals to me, tbh.

.02 euro per GB for premium Bandwidth + data storage. Compared to S3 Europe that is ridiculously good if you can eat the slightly higher latency.

https://www.runabove.com/storage/object-storage.xml

AMD, Applied Micro, and Cavium all have ARM 64-bit SOC's due out in servers in 2H2015, from a variety of OEM's. These SOC chips are designed for servers, with ECC, integrated SATA, 10Gbe, network accelerators, etc.

In the mean-time, 64-bit multi-core ARM dev boards are becoming available via www.96boards.org. Price for these is around $100 each instead of the $3K early access reference boards with the the server grade chips. Expectation is that 96boards.org will be offering boards with server grade chips in 2H2015 to further spur the SW ecosystem.

I mean, obviously, that isn't a realistic price, but if you want an ARM server with ECC ram right now, that's one way to go. (my understanding is that once AMD ramps up production, which is looking less likely at this point, it will be on the order of a few hundred bucks.)

Most people actually doing things with ARM servers are using non-ecc ram, which I don't think is particularly practical for real world use. The problem with non-ecc is not that it fails more often; the problem with non-ecc is that you often don't know when it failed.

Why is that? What changed?

I mean, from a shallow evaluation, I think that if these things came out at the "couple hundred dollar" price point, they would be competitive right now. at $300 per for board+cpu, I'd buy 20 right now. But I'm not at all sure that will be true a year from now.

(Note, I'm entirely ecc, so we're not talking about atoms. There is a huge opportunity here because intel stunts it's e3 line to 32gib ram. AMD needs to get in and exploit this while the getting is good. Yes, the 4xxx CPUs don't have any such limit, but compared to the E3, they are pathetic. These ARM boards would give AMD a chance to offer a reasonable compute per watt in a package with a ram footprint that can crush the E3.)

But... it's a year and a half later, and I still see no sign of AMD actually selling the things, other than as engineering samples at engineering sample prices, and meanwhile, intel is improving their x86 lineup.

Maybe I'm just impatient, but I just don't have a lot of hope that by the time it is out it will be able to compete with virtualizing a big xeon box.

I mean, yes, there are some applications where you really want your own server, and the A1100, if it does ship before Intel releases the ram limits on the E3, could compete in that niche, but for ARM to really take off in the datacenter, it needs to be cheaper per unit work done than a big xeon box. E.g. I need to be able to replace one of my giant dual xeons with four or five of these puppies, and it needs to cost less over the life of the hardware.

I'm paying California prices for power, so it shouldn't be that hard, but I'm just really frustrated by the slow progress of the AMD ARM server (and ARM boards that support ECC in general.)

http://b2b.gigabyte.com/products/product-page.aspx?pid=5423#...

Wondering if they'll roll-out in more regions soon. Interesting to see the international take-up, of course quite an euro-centric approach :)

"Deploy dedicated SSD servers with constant and predictable performances (sic) in 44 seconds."

More typos as you scroll down.

fwiw, i am from the east coast us, and have used both online.net labs and ovh dedicated cloud from ks and sys, and I have found that online.net has a lower latency and higher throughput to me.

also fwiw, i have not had issues with finding arm packages for developing in node, golang, or mono

There's a blog post about it at [1]

--

[0] http://labs.online.net/ [1] https://blog.scaleway.com/2015/04/02/from-online-labs-to-sca...

c1 from instantcloud.io:

  ubuntu@c1-10-1-33-39:~$ openssl speed rsa
  Doing 512 bit private rsa's for 10s: 11119 512 bit private RSA's in 10.00s
  Doing 512 bit public rsa's for 10s: 111478 512 bit public RSA's in 10.00s
  Doing 1024 bit private rsa's for 10s: 1950 1024 bit private RSA's in 10.00s
  Doing 1024 bit public rsa's for 10s: 34638 1024 bit public RSA's in 10.00s
  Doing 2048 bit private rsa's for 10s: 291 2048 bit private RSA's in 10.01s
  Doing 2048 bit public rsa's for 10s: 9393 2048 bit public RSA's in 10.00s
  Doing 4096 bit private rsa's for 10s: 40 4096 bit private RSA's in 10.16s
  Doing 4096 bit public rsa's for 10s: 2402 4096 bit public RSA's in 10.01s
  OpenSSL 1.0.1f 6 Jan 2014
  built on: Thu Oct 16 16:07:24 UTC 2014
  options:bn(64,32) rc4(ptr,char) des(idx,cisc,16,long) aes(partial) blowfish(ptr)
  compiler: cc -fPIC -DOPENSSL_PIC -DOPENSSL_THREADS -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -DL_ENDIAN -DTERMIO -g -O2 -fstack-protector-strong -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -Wl,-Bsymbolic-functions -Wl,-z,relro -Wa,--noexecstack -Wall -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_GF2m -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM -DAES_ASM -DGHASH_ASM
                    sign    verify    sign/s verify/s
  rsa  512 bits 0.000899s 0.000090s   1111.9  11147.8
  rsa 1024 bits 0.005128s 0.000289s    195.0   3463.8
  rsa 2048 bits 0.034399s 0.001065s     29.1    939.3
  rsa 4096 bits 0.254000s 0.004167s      3.9    240.0

  thresh@odroid:~$ openssl speed rsa
  Doing 512 bit private rsa's for 10s: 19334 512 bit private RSA's in 9.99s
  Doing 512 bit public rsa's for 10s: 204957 512 bit public RSA's in 9.97s
  Doing 1024 bit private rsa's for 10s: 3490 1024 bit private RSA's in 10.00s
  Doing 1024 bit public rsa's for 10s: 62453 1024 bit public RSA's in 10.00s
  Doing 2048 bit private rsa's for 10s: 516 2048 bit private RSA's in 10.00s
  Doing 2048 bit public rsa's for 10s: 17051 2048 bit public RSA's in 10.00s
  Doing 4096 bit private rsa's for 10s: 72 4096 bit private RSA's in 10.09s
  Doing 4096 bit public rsa's for 10s: 4434 4096 bit public RSA's in 10.00s
  OpenSSL 1.0.1f 6 Jan 2014
  built on: Fri Jun 20 19:00:28 UTC 2014
  options:bn(64,32) rc4(ptr,char) des(idx,cisc,16,long) aes(partial) blowfish(ptr)
  compiler: cc -fPIC -DOPENSSL_PIC -DOPENSSL_THREADS -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -DL_ENDIAN -DTERMIO -g -O2 -fstack-protector --param=ssp-buffer-size=4 -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -Wl,-Bsymbolic-functions -Wl,-z,relro -Wa,--noexecstack -Wall -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_GF2m -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM -DAES_ASM -DGHASH_ASM
                    sign    verify    sign/s verify/s
  rsa  512 bits 0.000517s 0.000049s   1935.3  20557.4
  rsa 1024 bits 0.002865s 0.000160s    349.0   6245.3
  rsa 2048 bits 0.019380s 0.000586s     51.6   1705.1
  rsa 4096 bits 0.140139s 0.002255s      7.1    443.4

  thresh@thre ~ $ openssl speed rsa
  Doing 512 bit private rsa's for 10s: 119305 512 bit private RSA's in 9.81s
  Doing 512 bit public rsa's for 10s: 1450435 512 bit public RSA's in 9.85s
  Doing 1024 bit private rsa's for 10s: 35347 1024 bit private RSA's in 9.85s
  Doing 1024 bit public rsa's for 10s: 541458 1024 bit public RSA's in 9.86s
  Doing 2048 bit private rsa's for 10s: 5259 2048 bit private RSA's in 9.84s
  Doing 2048 bit public rsa's for 10s: 161094 2048 bit public RSA's in 9.88s
  Doing 4096 bit private rsa's for 10s: 627 4096 bit private RSA's in 9.81s
  Doing 4096 bit public rsa's for 10s: 41353 4096 bit public RSA's in 9.75s
  OpenSSL 1.0.1e 11 Feb 2013
  built on: Thu Mar 19 18:31:36 UTC 2015
  options:bn(64,64) rc4(16x,int) des(idx,cisc,16,int) aes(partial) blowfish(idx)
  compiler: gcc -fPIC -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -m64 -DL_ENDIAN -DTERMIO -g -O2 -fstack-protector --param=ssp-buffer-size=4 -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -Wl,-z,relro -Wa,--noexecstack -Wall -DMD32_REG_T=int -DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_MONT5 -DOPENSSL_BN_ASM_GF2m -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DVPAES_ASM -DBSAES_ASM -DWHIRLPOOL_ASM -DGHASH_ASM
                    sign    verify    sign/s verify/s
  rsa  512 bits 0.000082s 0.000007s  12161.6 147252.3
  rsa 1024 bits 0.000279s 0.000018s   3588.5  54914.6
  rsa 2048 bits 0.001871s 0.000061s    534.5  16305.1
  rsa 4096 bits 0.015646s 0.000236s     63.9   4241.3

  [ec2-user@ip-10-71-147-165 ~]$ openssl speed rsa
  Doing 512 bit private rsa's for 10s: 85447 512 bit private RSA's in 9.48s
  Doing 512 bit public rsa's for 10s: 123882 512 bit public RSA's in 9.96s
  Doing 1024 bit private rsa's for 10s: 7720 1024 bit private RSA's in 9.98s
  Doing 1024 bit public rsa's for 10s: 187876 1024 bit public RSA's in 10.04s
  Doing 2048 bit private rsa's for 10s: 1268 2048 bit private RSA's in 9.99s
  Doing 2048 bit public rsa's for 10s: 67337 2048 bit public RSA's in 10.07s
  Doing 4096 bit private rsa's for 10s: 36 4096 bit private RSA's in 7.89s
  Doing 4096 bit public rsa's for 10s: 25180 4096 bit public RSA's in 9.85s
  OpenSSL 1.0.1g-fips 7 Apr 2014
  built on: Wed Apr  9 21:51:31 UTC 2014
  options:bn(64,64) md2(int) rc4(16x,int) des(idx,cisc,16,int) aes(partial) idea(int) blowfish(idx)
  compiler: gcc -fPIC -DOPENSSL_PIC -DZLIB -DOPENSSL_THREADS -D_REENTRANT -DDSO_DLFCN -DHAVE_DLFCN_H -DKRB5_MIT -m64 -DL_ENDIAN -DTERMIO -Wall -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector --param=ssp-buffer-size=4 -m64 -mtune=generic -Wa,--noexecstack -DPURIFY -DOPENSSL_IA32_SSE2 -DOPENSSL_BN_ASM_MONT -DOPENSSL_BN_ASM_MONT5 -DOPENSSL_BN_ASM_GF2m -DSHA1_ASM -DSHA256_ASM -DSHA512_ASM -DMD5_ASM -DAES_ASM -DVPAES_ASM -DBSAES_ASM -DWHIRLPOOL_ASM -DGHASH_ASM
                    sign    verify    sign/s verify/s
  rsa  512 bits 0.000111s 0.000080s   9013.4  12438.0
  rsa 1024 bits 0.001293s 0.000053s    773.5  18712.7
  rsa 2048 bits 0.007879s 0.000150s    126.9   6686.9
  rsa 4096 bits 0.219167s 0.000391s      4.6   2556.3

t1.micro running ubuntu

  sign                                 verify         sign/s       verify/s
  rsa  512 bits 0.000104s 0.000015s   9647.3  67715.6
  rsa 1024 bits 0.003637s 0.000236s    275.0   4242.7
  rsa 2048 bits 0.004487s 0.000107s    222.8   9351.2
  rsa 4096 bits 0.180566s 0.003013s      5.5    331.9

  sign    verify    sign/s verify/s
  rsa  512 bits 0.000109s 0.000009s   9190.4 114333.6
  rsa 1024 bits 0.000354s 0.000023s   2822.7  42725.7
  rsa 2048 bits 0.002491s 0.000078s    401.5  12833.4
  rsa 4096 bits 0.017951s 0.000289s     55.7   3454.7

                  sign    verify    sign/s verify/s
  rsa  512 bits 0.000060s 0.000005s  16569.7 208219.9
  rsa 1024 bits 0.000196s 0.000013s   5113.5  74559.9
  rsa 2048 bits 0.001459s 0.000045s    685.3  22400.0
  rsa 4096 bits 0.010396s 0.000165s     96.2   6050.9

# openssl speed rsa -multi 8

...

                  sign    verify    sign/s verify/s

Support ticket answered in < 2 minutes. Resolved in 9 minutes.

(The ticket was sort of my fault, partly a design issue on their UI where I couldn't resubmit info. There are teething problems, but they sorted it out really quickly.)

  ARMv7 Processor rev 2 (v7l) (4) 1.0.0-SNAPSHOT
  min time: 4
  max time: 1655
  mean time: 276.8
  median time: 148.5

  Intel(R) Xeon(R) CPU E5-2630L v2 @ 2.40GHz (1) 1.0.0- SNAPSHOT
  min time: 1
  max time: 261
  mean time: 51.0447368421
  median time: 30.0

  Intel(R) Xeon(R) CPU E5-2630L v2 @ 2.40GHz (2) 1.0.0-SNAPSHOT
  min time: 1
  max time: 407
  mean time: 64.6447368421
  median time: 34.0

  Intel(R) Core(TM) i5-3570K CPU @ 3.40GHz (4) 1.0.0-SNAPSHOT
  min time: 0.0001
  max time: 114
  mean time: 21.9026331579
  median time: 12.0

  Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz (2) 1.0.0-SNAPSHOT
  min time: 5
  max time: 7273
  mean time: 417.507894737
  median time: 103.5

I was impressed with ARM speed (it is much better then free openshift trial maybe because it only has 512MB) but now after free trial I'll probably move to DO the cheapest package, because for 11€ I can get better performance on DO even though it is varied. During normal use I can see the difference in speed between ARM and openshift in requests but not so much between ARM and DO.

Probably ARM would work better when multiple request would come (It has 4 cores) and DO has 1/2.

Their "Infinite Storage" is €0.02/month, though.

[0] https://www.scaleway.com/pricing

Each resource is billed per hour up to its monthly cap. The monthly cost of a resource is calculated by running the hourly rate for 500 hours. If a resource is used during more than 500 continuous hours in the month, the rest of the month is free."

"Infinite Storage" is, like the term "unlimited", misleading. I clicked through only to find out that it's for 1GB of storage, with unlimited requests and transfers.

It is almost assuredly a lie (as someone who is working on a project scaling to PBs -- I pressed a lot of vendors on "unlimited" or "infinite" and most said they simply couldn't scale to X).

And the term "infinite" is misleading, as it's actually just 1GB of data that you can transfer as much as you want.

Unlimited transfer seems too good to be true.

Just one problem, I'm getting an internal server error while verifying my address (home address), and the SMS isn't arriving. I'll try it a little later. But I am a little confused, how exactly do you verify a home address?

doc: https://origin-www.marvell.com/embedded-processors/armada-xp...

Do you know the latency numbers?

Edit: found this https://www.ruhr-uni-bochum.de/integriertesysteme/emuco/file... (slide 17)

"How does a C1 server perform? A C1 server gives a constant CPUMark of 12K. This is equivalent to an AWS M3 medium instance. "

Edit: It seems that ARM CPUs got really fast: According to [1] an Intel Core i7-3930K @ 3.20GHz gets a Passmark score of 12k.

[1]: https://www.cpubenchmark.net/high_end_cpus.html

Perhaps they are using the Android benchmark. A score of 12,000 is about what a quad core ARMv7 phone gets.

http://www.androidbenchmark.net/cpumark_chart.html

Edit:

Compared with the cross-platform Geekbench3 a HTC One (ARMv7 4 core, 1.7-2.2Ghz, 12,000 Android CPUMarks) gets between 2,000 and 3,000 points[1], while a 5820K gets around 20,000 points[2].

[1]: http://browser.primatelabs.com/geekbench3/search?q=HTC+One

[2]: http://browser.primatelabs.com/geekbench3/search?q=Intel+Cor...

http://browser.primatelabs.com/geekbench3/search?dir=desc&q=...

Glad to see they support Docker on them, that's one of the killer apps for ARM systems.

I rather pay $40/mo for Hetzner 3930 CPU server with something like 30 VMs.

It's no great saving over places like Hetzner or Digital Ocean, but I don't think it's _quite_ as bad a deal as you're making out... I wonder how much more than $300 it'd cost you to buy the hardware that Hetzner give you for four times as much per month?

Otherwise, someone willing to pay as little as possible for just a single bare metal dedicated machine, is much better of renting a Kimsufi instead.

http://www.kimsufi.com/fr/

Raspberry Pis and SSDs do not come with bandwidth and redundant network, power, and cooling infrastructure.

Good luck when Hetzner null route your server when a competitor DDoSes you

Just look on the various web hosting forums (webhostingtalk.com) with regards to Hetzer. They are very hostile to customers that have these problems.

OVH, which are priced similarly, include DDoS protection in all their new plans.

I find Hetzer are good for non public facing servers. Job processors, internal tool servers etc.

For the best protection you need to pay more - in the case of Scaleway, their pricing page (https://www.scaleway.com/pricing) says it's available with the "premium" bandwidth option - but anecdotally I've heard even the basic protection is pretty good.

Is this really not the case? In 2015?

Some of the things I might do with these don't really have that "better than the x86 version" comparison in any rel sense. I've got a DigitalOcean droplet and a CloudAtCost VM (and several others) that exclusively get used as remote endpoints for reverse SSH tunnels. These would only need to be slightly less expensive for me to seriously consider adding a French jurisdiction one as well...

https://hub.scaleway.com/

"Online SAS, a company registered in France."

No imprint in France? Illegal.

    ONLINE, a simplified stock corporation (Société anonyme par actions simplifiée) with a working capital
    of €214.410,50, headquartered at 8 rue de la ville l'Evêque - 75008 Paris, FRANCE, registered with the
    Paris Corporate and Trade Register number RCS PARIS B 433 115 904, VAT number FR35433115904, reachable
    via its Internet site http//www.scaleway.com/ as well as by telephone at +33 (0) 184 130 000", or by 
    fax at +33 (0) 899 193 775 (€1.35 per call plus €0.34/min.)

The costs seemed very high, to me, compared to whitebox/barebones AMD or even Intel based servers on a performance basis.

I think it is simply a function of them not being widely available cheaply yet. I wonder how they're getting them cheap enough to make this make sense...maybe quantity pricing is much better.

Memory capacity was actually my biggest cost factor that made a difference. You can buy x86 boxes with 256GB capacity for quite cheap. Up to 1TB for not outrageous amounts. The micro servers are designed for quite low amounts of memory... But memory is, by far, the bottleneck I am concerned with. CPU speed barely matters at all...so virtualizing CPU is fine as long as memory can be huge and fast.

> Scaleway is also providing a fast, 100% S3-compatible, object storage service called SIS (Scaleway Infinit Storage). SIS starts at €0.00004 for 1GB of storage.

Should be 'infinite'?

> At Scaleway, we provide bare metal SSD servers. When you request a server, you get full root access on a dedicated servers.

Plural agreement. Should be 'on a dedicated server'.

> A simple, robust REST APIs to control all your cloud ressources

Plural agreement, 'Simple, robust REST APIs'. Also spelling, 'resources'.

4-cores ARMv7 CPU with 2GB of RAM and a 1Gbit/s network card

source: https://blog.scaleway.com/2015/04/02/from-online-labs-to-sca...

They have only one OS available, Linux, (in various flavours), while the others are branded 'coming soon'. That alone doesn't inspire any confidence.

In short, while its cool if they offer other OS options, my confidence certainly is not shaken by them choosing to roll out first with just Linux support. It seems like the only sensible thing, to me, given how good Linux ARM support is.

OpenBSD. Or any of the BSDs, but OpenBSD would be ideal.

That said, my confidence isn't exactly shaken either by the current lack of BSD availability (FreeBSD is supposedly "coming soon"). I'm getting by moderately well on Linux right now for my own "cloud" needs. It would just be nice to have my preference in operating systme (and ecosystem; I've taken a liking to OpenBSD's subprojects like relayd/httpd, OpenSMTPd, etc., and it's much easier to use these things on OpenBSD - or probably even on FreeBSD - than it is on GNU/Linux).