Scaleway ARMv8 Cloud Servers

Scaleway can &quot;disrupt&quot; the industry as much as they want. As long as they don&#x27;t have a solid product it&#x27;s useless for production usage (backups, weird storage system, unpredictable network, etc.)<p>They should just offer cheap Xeon D VMs and not this ARM stuff. The CPU is way too slow and the ratio is completely off, 2 of these ARM cores for 2 GB RAM? Even a shared Xeon or Xeon D core offers better performance than those ARM CPUs.

I have a $3 ARMv7 Scaleway server, and I can vouch for the unlimited bandwidth. I&#x27;ve run tens of terabytes (in a single month) through it as a torrent mirror for various Linux ISO images.<p>The disk speed is a bit slow, and 200mbits is a bit low, but I still wish they had a US data center so I can use them as a fail-over host.

Great pricing, but literally no customer support. Week&#x27;s ago I wanted to change my billing address, I was surprised not to find the option within my account settings, so I asked customer support to do so, whom replied that they simply can&#x27;t do it.<p>I can just close my current account, then recreate it with my new address and of course, there is no migration tools.

There are a lot of comments in here about the CPU&#x2F;RAM but I think the important thing (at least to me) is that you can obtain 200Mbit&#x2F;s for 3EUR&#x2F;mo. Saturated, that&#x27;s 65TB&#x2F;mo, which works out to 0.00005 $&#x2F;GB. Most of the big clouds and CDNs charge on the order of 0.08 $&#x2F;GB. Sure you can get unlimited bandwidth on some of the bigger dedicated server packages but not for 3EUR&#x2F;mo.<p>And yes, all networks are not equal. 65TB on a shitty network isn&#x27;t worth as much if it has 800ms latency or something like that but at least there&#x27;s an option.

I am super excited for this. I&#x27;ve been running an ARMv7 server on Scaleway almost since they were released.<p>However, I don&#x27;t see any news that the new ARMv8 servers will support native IPv6.<p>This is a big limitation, if you want to be IPv6 accessible now, you have to run a 6in4 tunnel. The x86 servers (physical and VM) from Scaleway all support native IPv6. The explanation so far was that the ARM back plane doesn&#x27;t support IPv6.<p>So will these new ARMv8 servers support native IPv6? Or are they build on the same back plane as the ARMv7 servers?<p>Edit: It&#x27;s supported, I can&#x27;t read.

Can anyone explain to an outsider why this is disruptive? Are these significantly faster than the old x64 offerings? A lot cheaper?<p>For that matter, has anyone done a real-world armv8 vs. x64 benchmark?<p>--<p>edit: would like to test them myself, unfortunately they only allow payment through a &quot;physical&quot; credit card: <a href="https:&#x2F;&#x2F;www.scaleway.com&#x2F;faq&#x2F;billing&#x2F;#-What-forms-of-payment-do-you-accept" rel="nofollow">https:&#x2F;&#x2F;www.scaleway.com&#x2F;faq&#x2F;billing&#x2F;#-What-forms-of-payment...</a>

I&#x27;ve played around with it. These CPUs are fast enough (certainly faster than Heroku and the like). However, ScaleWay still does not provide automated disk backups, and manual snapshot requires you to shut down the instance first, so I went back to Linode.

Let&#x27;s challenge ourselves to not &quot;disrupt&quot;, or &quot;make the world a better place&quot; in 2017. I&#x27;m trying to stem the flow of ribbing we will get 20 years down the line from our peers for using these types of marketing tropes.

There is no ImageHub for the new ARM servers - the only images offered are Ubuntu and Jessie, i.e. no readymade Docker setup.<p>Maybe that is obvious, but the new servers can&#x27;t run Raspberry Pi3 Ubuntu binaries. The old ARM servers also have issues with that because the Scaleway ARM7 CPUs lacks features provided by the Raspi3 ARM CPUs (e.g. they lack Neon).<p>Summary: This is not a hosted Raspi replacement :-) But still quite nice.

Full-stack dev here with almost no real hardware knowledge.<p>Why would you want to use ARM instead of X86 on a server like this? The architecture​ switch comes with dependency headaches -- what counterbalances those to make it worth it?

<p><pre><code> We currently offer Ubuntu 16.04 as base Operating System </code></pre> I see a lot of providers offering Ubuntu these days. Is Ubuntu meanwhile more popular with providers then Debian? And why did it become so popular? Does it have any technical advantages over Debian or is it because many developers have it on their local machine and therefore want the same OS on the server?

Note that Scaleway bills in euros, not dollars, pounds, etc. They also have no pre-paid policy, so you cannot pay a lump sum upfront to get many months usage. This means that you could be hit by significant monthly fees if you are paying via another currency. But of course, different banks have different policies, and YMMV.<p>My €3&#x2F;mo was (IIRC) more like €5&#x2F;mo once conversion fees were taken into account. This €2 flat fee would not have been a problem had I been paying for 12 months at a time, but Scaleway do not offer this, in contrast to many other hosting providers on the market. After inquiring, I was told there was no current plan to change this policy.

Reminded me of this article &quot;Cellphone Chips Will Remake the Server World. Period.&quot; [1] from 2013.<p>And also of SeaMicro [2].<p>Looks like the wimpy core lost this particular battle. Or maybe they only make sense at a particular scale.<p>[1] <a href="https:&#x2F;&#x2F;www.wired.com&#x2F;2013&#x2F;01&#x2F;facebook-arm-chips&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.wired.com&#x2F;2013&#x2F;01&#x2F;facebook-arm-chips&#x2F;</a><p>[2] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;SeaMicro" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;SeaMicro</a>

These seem very inexpensive for the memory and storage, irrelevant of how powerful the processor is. Looks like they&#x27;d make great little VPN servers.

8GB RAM, 200GB SSD for $11.99 is significantly cheaper than linode&#x27;s 8GB RAM 96GB SSD for $40. Then again all of linode&#x27;s plans are cheaper than Digital Ocean&#x27;s. I am curious if this price gap is due to the cheaper ARM SoC ?

I have a bunch of Windows Server key&#x27;s through Bizspark, Scaleway was the cheapest option to do hosted Windows Servers ( 5 $ &#x2F; month). It takes some effort though ( adding scaleway drivers to a windows server ISO)<p>Awesome

Little bit confusing in the admin interface. It seems like these new processors replace the old C1 offer. This is the cpuinfo of the new one:<p><pre><code> cat &#x2F;proc&#x2F;cpuinfo processor : 0 BogoMIPS : 200.00 Features : fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics CPU implementer : 0x43 CPU architecture: 8 CPU variant : 0x1 CPU part : 0x0a1 CPU revision : 1 processor : 1 BogoMIPS : 200.00 Features : fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics CPU implementer : 0x43 CPU architecture: 8 CPU variant : 0x1 CPU part : 0x0a1 </code></pre> For comparison, this was the old C1, their prior ARM offering they mention in the blog post:<p><pre><code> Processor : Marvell PJ4Bv7 Processor rev 2 (v7l) processor : 0 BogoMIPS : 1332.01 processor : 1 BogoMIPS : 1332.01 processor : 2 BogoMIPS : 1332.01 processor : 3 BogoMIPS : 1332.01 Features : swp half thumb fastmult vfp edsp vfpv3 tls CPU implementer : 0x56 CPU architecture: 7 CPU variant : 0x2 CPU part : 0x584 CPU revision : 2 Hardware : Online Labs C1 Revision : 0000 Serial : 0000000000000000 </code></pre> I take it the BogoMips are just a false info in the kernel interface.<p>For performance, I ran `sysbench --test=cpu --cpu-max-prime=1000 run` on both (first I tried that with max-prime=20000, but that did not finish in a reasonable time span on the old server).<p>New:<p><pre><code> Number of threads: 1 Doing CPU performance benchmark Threads started! Done. Maximum prime number checked in CPU test: 1000 Test execution summary: total time: 0.6541s total number of events: 10000 total time taken by event execution: 0.6507 per-request statistics: min: 0.06ms avg: 0.07ms max: 0.14ms approx. 95 percentile: 0.06ms Threads fairness: events (avg&#x2F;stddev): 10000.0000&#x2F;0.00 execution time (avg&#x2F;stddev): 0.6507&#x2F;0.00 </code></pre> Old:<p><pre><code> Number of threads: 1 Doing CPU performance benchmark Threads started! Done. Maximum prime number checked in CPU test: 1000 Test execution summary: total time: 10.1997s total number of events: 10000 total time taken by event execution: 10.1909 per-request statistics: min: 1.02ms avg: 1.02ms max: 1.22ms approx. 95 percentile: 1.02ms Threads fairness: events (avg&#x2F;stddev): 10000.0000&#x2F;0.00 execution time (avg&#x2F;stddev): 10.1909&#x2F;0.00 </code></pre> When using the additional cores (2 for the new, 4 for the old):<p>New:<p><pre><code> Number of threads: 2 Doing CPU performance benchmark Threads started! Done. Maximum prime number checked in CPU test: 1000 Test execution summary: total time: 0.3282s total number of events: 10000 total time taken by event execution: 0.6523 per-request statistics: min: 0.06ms avg: 0.07ms max: 0.28ms approx. 95 percentile: 0.06ms Threads fairness: events (avg&#x2F;stddev): 5000.0000&#x2F;2.00 execution time (avg&#x2F;stddev): 0.3261&#x2F;0.00 </code></pre> Old:<p><pre><code> Number of threads: 4 Doing CPU performance benchmark Threads started! Done. Maximum prime number checked in CPU test: 1000 Test execution summary: total time: 2.5527s total number of events: 10000 total time taken by event execution: 10.1983 per-request statistics: min: 1.02ms avg: 1.02ms max: 5.49ms approx. 95 percentile: 1.02ms Threads fairness: events (avg&#x2F;stddev): 2500.0000&#x2F;1.87 execution time (avg&#x2F;stddev): 2.5496&#x2F;0.00 </code></pre> The new processors seem to be a lot more powerful. I run my blog on that old C1 (that might have skewed results somewhat if they were close, but they are not). I think I&#x27;ll move it over asap.<p><i>Edit:</i><p>I have another Scaleway instance that runs <a href="https:&#x2F;&#x2F;www.pc-kombo.de" rel="nofollow">https:&#x2F;&#x2F;www.pc-kombo.de</a>, and is a C2S instance, which is the also mentioned X64 offering of theirs. It is more expensive, as it has more cores, but if matching those the ARMv8 seems to actually be faster. That&#x27;s the performance of the C2S:<p><pre><code> Number of threads: 1 Doing CPU performance benchmark Threads started! Done. Maximum prime number checked in CPU test: 1000 Test execution summary: total time: 0.8497s total number of events: 10000 total time taken by event execution: 0.8480 per-request statistics: min: 0.08ms avg: 0.08ms max: 0.13ms approx. 95 percentile: 0.08ms Threads fairness: events (avg&#x2F;stddev): 10000.0000&#x2F;0.00 execution time (avg&#x2F;stddev): 0.8480&#x2F;0.00 </code></pre> With 2 threads:<p><pre><code> Number of threads: 2 Doing CPU performance benchmark Threads started! Done. Maximum prime number checked in CPU test: 1000 Test execution summary: total time: 0.4489s total number of events: 10000 total time taken by event execution: 0.8752 per-request statistics: min: 0.08ms avg: 0.09ms max: 20.09ms approx. 95 percentile: 0.08ms Threads fairness: events (avg&#x2F;stddev): 5000.0000&#x2F;23.00 execution time (avg&#x2F;stddev): 0.4376&#x2F;0.01 </code></pre> Seems like I should move that as well.

These arm servers, do they use ECC ram?

Slightly OT but can anyone recommend ARM servers for home use? I have x86 server but is too noisy and overpowered, and a raspberry pi which is too weak. Can someone recommend some hardware in-between?

That&#x27;s perfect to run Tor relay services on the cheap

they still lack two factor auth...