ARMv7 dedicated servers for 3.40 per month
techcrunch.comEarlier discussion here: https://news.ycombinator.com/item?id=10159989
I've just signed up for one of these instances - here are some speed tests/stats/details, if anyone's interested: https://gist.github.com/JosephRedfern/57c81ac95bf4f1175e3e
If you'd like anything in particular running, let me know and I shall oblige when I can!
Hi, can you please run the following two fio tests:
#Random write fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=128 --size=4G --readwrite=randwrite #Random Read fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=128 --size=4G --readwrite=randread~# fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=16 --size=4G --readwrite=randread test: (groupid=0, jobs=1): err= 0: pid=17457: Thu Sep 3 12:39:10 2015 read : io=0B, bw=115612KB/s, iops=28903, runt= 36279msec cpu : usr=17.12%, sys=67.12%, ctx=35506, majf=0, minf=39 IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=100.0%, 32=0.0%, >=64=0.0% submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.1%, 32=0.0%, 64=0.0%, >=64=0.0% issued : total=r=1048576/w=0/d=0, short=r=0/w=0/d=0 Run status group 0 (all jobs): READ: io=4096.0MB, aggrb=115612KB/s, minb=115612KB/s, maxb=115612KB/s, mint=36279msec, maxt=36279msec Disk stats (read/write): nbd0: ios=55484/53, merge=0/587, ticks=134950/1020, in_queue=135900, util=24.02% ~# fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=16 --size=4G --readwrite=randwrite test: (groupid=0, jobs=1): err= 0: pid=17460: Thu Sep 3 12:43:54 2015 write: io=0B, bw=15741KB/s, iops=3935, runt=266460msec cpu : usr=3.97%, sys=25.75%, ctx=544104, majf=0, minf=23 IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=100.0%, 32=0.0%, >=64=0.0% submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.1%, 32=0.0%, 64=0.0%, >=64=0.0% issued : total=r=0/w=1048576/d=0, short=r=0/w=0/d=0 Run status group 0 (all jobs): WRITE: io=4096.0MB, aggrb=15740KB/s, minb=15740KB/s, maxb=15740KB/s, mint=266460msec, maxt=266460msec Disk stats (read/write): nbd0: ios=0/1053524, merge=0/106948, ticks=0/4282740, in_queue=4281210, util=100.00% ~# fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=128 --size=4G --readwrite=randwrite test: (groupid=0, jobs=1): err= 0: pid=17463: Thu Sep 3 12:48:26 2015 write: io=0B, bw=16400KB/s, iops=4100, runt=255749msec cpu : usr=4.20%, sys=22.28%, ctx=576266, majf=0, minf=23 IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0% submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.1% issued : total=r=0/w=1048576/d=0, short=r=0/w=0/d=0 Run status group 0 (all jobs): WRITE: io=4096.0MB, aggrb=16400KB/s, minb=16400KB/s, maxb=16400KB/s, mint=255749msec, maxt=255749msec Disk stats (read/write): nbd0: ios=0/1047606, merge=0/56, ticks=0/32543280, in_queue=32545150, util=100.00% ~# fio --randrepeat=1 --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=test --bs=4k --iodepth=128 --size=4G --readwrite=randread test: (groupid=0, jobs=1): err= 0: pid=9439: Thu Sep 3 12:48:32 2015 read : io=0B, bw=28594KB/s, iops=7148, runt=146685msec cpu : usr=8.60%, sys=31.77%, ctx=591701, majf=0, minf=138 IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0% submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.1% issued : total=r=1048576/w=0/d=0, short=r=0/w=0/d=0 latency : target=0, window=0, percentile=100.00%, depth=128 Run status group 0 (all jobs): READ: io=4096.0MB, aggrb=28593KB/s, minb=28593KB/s, maxb=28593KB/s, mint=146685msec, maxt=146685msec Disk stats (read/write): nbd0: ios=1047291/4, merge=0/1, ticks=18660030/50, in_queue=18659150, util=100.00%Officially they provide 2000 read/write IOPS for a volume composed of one chunk (50GB) [1]. You measured 4000 r/w IOPS, so does this mean you ran it on a 100GB volume? If yes, then that's pretty amazing you get exactly the expected allocated IOPS performance for a $3.40/mo server.
I'm on a 50GB volume :)
2000 IOP/s is pretty awful, that's 50x slower than even a single SATA SSD such as a Sandisk Extreme Pro.
Can confirm that my 50GB box is giving almost identical numbers.
Sure - I'll get back to you in a couple of hours - I'm working right now, I'm afraid.
Thanks, no rush, would be interesting to see. Off to sleep so I'll reply back in the morning.
Is it possible to physically buy these servers? All the info indicates that you can just get instances, just like the rest of the server farms.
You're right - rephrased :)
Do you think it is possible to run the android emulator on this server? Or will it be an issue because it's running on armhf?
Looks like Android support is on the coming soon list, you can vote it up https://www.scaleway.com/imagehub/
> If you'd like anything in particular running, let me know and I shall oblige when I can!
bonnie++ would be great
Thanks for taking time to do this.
Nitpicking the submission title ("ARM7 dedicated servers for 3.40 per month"): ARM7 is actually very old ARM cores running ARMv4; these servers are ARMv7.
To be even more specific, these CPU are produced by Marvell under the name Armada 370/XP (according /proc/cpuinfo), and are based on ARM's Cortex A9 design (source: https://scaleway.com/faq/server/#-Why-did-you-choose-ARM ).
According to /proc/cpuinfo, CPU Implementer is 0x56 ('V' = Marvell) and CPU part is 0x584 (Marvell Armada XP). This is most likely not Cortex A9 (which would have CPU Implementer == 0x41 and CPU part == 0xC09), but Marvell's own PJ4 microarchitecture.
Maybe yes. I'm not part of the hardware team so I can't really tell. Was just quoting the FAQ here.
Fixed this up, sorry, don't have a lot of knowledge about ARM architecture.
Tangential rant: What is wrong with CPU manufacturers in particular?
What do they gain from giving everything the most confusing names possible?
Sockets LGA1155, LGA1150, LGA1151. CPUs only differentiated by a jumble of digits anymore.
Something is seriously wrong with your product naming strategy when it needs to be clarified that ARM7 != ARMv7.
Intel uses the letters to indicate the type of connection between the CPU and motherboard. So LGA = Land Grid Array [1]. The number is the number of pins/contacts connecting the CPU to the motherboard. So LGA1155 is a Land Grid Array Socket with 1,155 pins.
It's definitely not random, at least not with Intel. ARM is confusing as hell though. Part of the issue is that there is the ARM Instruction Set Architecture (ISA) that has different versions, and then there are different ARM CPU architectures, but those two don't necessarily need to correspond.
So LGA1155 is a Land Grid Array Socket with 1,155 pins.
Their scheme may be somewhat internally consistent (until they change it again next year). The information it conveys remains completely useless.
Nobody cares about the number of pins on a socket or which "generation" a CPU belongs to.
We want to know which CPU is compatible to which socket, and how CPUs compare on their key metrics.
So here's a useful naming scheme: Sockets should be called S1, S2, S3 [...]
CPUs should be called: S1-8-40W-PM15000
That would be an 8 core CPU for Socket S1 that draws 40W and scores PassMark 15000.
If Intel & co insist on making us sift through hostile numeric identifiers then they should at least make them useful.
Is everyone able to fix submission title on HN now or do you have some special status (e.g. mod or high karma)?
Well, he is the OP.
But what about bandwidth? That's still shared among all the users, right?
And given that they are even advertising 'Seedboxes' and VPN servers [0] hosted in the same network, it makes me doubt the quality of network even more.
[0] https://blog.scaleway.com/2015/09/02/we-are-slashing-the-c1-...
They have another product mini dedicated for 6 euro/mo[0]. Which I've been using recently and the network performance has been pretty consistent. They are however seem marketed more for personal usage though.
BW on these is shared a rack level, depending on which rack you're put in your speeds can vary a lot
The submission title really should include a dollar to indicate which currency is meant.
Also, for non-corporate customers in Europe, VAT should be added.
The VAT-inclusive price is €3.59 (for 4 ARM cores, 2 GB RAM, 50 GB SSB disk, and one IP with 200 Mbps unmetered bandwidth). This is the VAT-inclusive number from the site, which is using French VAT (20%). Here in Sweden it'd be 25%, it seems. :|
For european union based users without a valid EU VAT number, VAT must be added based on the VAT rate in the users own country, not always french VAT at 20%. Ie we pay 25% here in DK. This is due to the new rules that went into effect 1 January 2015: http://ec.europa.eu/taxation_customs/taxation/vat/how_vat_wo...
I concur. Rules and rates are summarized here: https://github.com/kdeldycke/vat-rates
Interestingly even though it's 'bare metal' you are stuck using their kernel: https://community.scaleway.com/t/official-linux-kernel-new-m...
/proc/cpuinfo hints at Armada XP so I thought mv_cesa crypto would be available but alas, no:
# zcat /proc/config.gz | grep -i cesa
CONFIG_CRYPTO_DEV_MV_CESA=y
# grep -i cesa /proc/crypto
#
edit: it's definitely marvell, dmesg references 370_xp clocksource, armada xp pincontrol and aurora L2 cache controller.
There was some discussion of that in the original announcement. Apparently the kernel requirement is because the root fs is a network drive, and the system PXE-boots: https://news.ycombinator.com/item?id=9309661
An interesting clause in the terms and conditions:
* Users recognise that piracy harms artistic creation.
I recognise that piracy harms artistic creation, but I don't care. So can I sign up? ;P
I am glad the shipping industry is no longer affected.
Likely a side effect of French laws.
Low cost hoster Hetzner also launched an ARM based dedicated server a few days ago: https://www.hetzner.de/us/hosting/produktmatrix/rootserver-p...
Kimsufi has been also providing cheapish dedicated servers (non-arm) for a long time: http://www.kimsufi.com/uk/
I wonder if I could buy a couple of those servers for my own. I work on cloud and data-center automation tools and having real metal to test ideas and software would be amazing. I've been toying with the idea of hooking up a couple Edisons or Galileos (x86 makes everything easier on my side) but they have complete running hardware and that easily beats developing my own.
BTW, the said software is open source, so there is even some good karma thrown in.
This was announced directly from Scaleway and was a pretty popular thread here on HN, why is an article about it on here again and just as popular?
I didn't click on the original article at the time because the title was utterly meaningless without context: "We are slashing the C1 price by 70 percent". I'm not motivated enough to figure out the context for every article, so I just skip those unless they have a very high number of points or comments. I'm probably not alone in this.
FYI - https://www.youtube.com/watch?v=XFhgSKNJP2s&feature=youtu.be
For my own projects, I wish I could buy this hardware!
I like the idea of the product a lot - but the question I have is how these servers stack up in an 'apples to apples' comparison (that is to say, with a DO small / medium / large instance - or an AWS small / medium / large instance) - which would allow for me to work out if it's worthwhile investing..
Is there any way to test out Scaleway without having to give them a credit card number with a potential infinite bill?
I did test them out a while ago. My impressions are here: https://www.amon.cx/blog/scaleway-vs-digitalocean-and-deploy...
You can try with http://instantcloud.io
You can try them for a short time for free, but you have to authorize via SMS.
There's also hard limits on the number of servers you can create (~10) until you contact support to get that value lifted so it appears there are caps on what you spend
I was going to sign up for this immediately. Then I read the FAQ:
- No IPv6 -- sucks, but I can live with that
- No IPv4 -- WTF? NAT on a server?!
- A credit card is required. I can't pay them. Asked via support, they confirm credit card is the only way.
- I still haven't found where the instances are physically located. (Edit: it's in the "General FAQ", not in the "Server FAQ".)
I really, really like the idea, but they have a long way to go.
The IPv4 address is independent of any one server. Think of it kind of like L2 / L3, when a packet gets to its final network the L3 address is translated to a L2 address (ARP typically with ethernet / IP) with the L3 frame wrapped by a L2 frame. So think of this more like L2 is now a combination of classic L2 and the lower end of L3. It's sort of halfway between home router style NAT and switching; and yes, I'm sure it runs as close to full speed as switching does. This kind of set up is very common in cloud based services that support live migration (not that this service does; it's bare metal).
They have made a couple of customizations to the regular Linux bootup that have severe security implications (i.e. download kernel modules from them among other things). Contrary to a "classic" dedicated server, with this configuration you are very much trusting them unless you make changes.
Read https://www.scaleway.com/docs/create-an-image-from-scratch/ and ideally set up a server and look at what exactly is going on.
Unfortunately the script on that page has some errors (it references missing files and it's missing some files that are required), they should really go over it and fix them.
Is the storage attached directly to the box or is it a SAN? I'd like to use one of these for a personal email server, but don't want to bother if a fried disk or disk controller would result in significant downtime for my email.
It's a SAN, mounted as a NBD.
No IPv6? This is a no-go for servers... Add it and i'm all in. ;)
A huge part of the monthly cost of any infrastructure is less about power and the hardware and more about the quality of the internet connection. If I can't get data in and out quickly it isn't going to be much good to me.
Scaleway is part of the Iliad group, the third French ISP: https://en.wikipedia.org/wiki/Iliad_SA
No need to worry about our infrastructure. Our datacenters are well connected: http://www.iliad-datacenter.fr/infrastructure/connectivite
Benchmark for those interested
their imagelab is pretty cool. finally a place to host my ghost blog cheaply.
Kudos to the folks at Scaleway, and mostly their engineering teams :)
Thanks ! :)
US data center please and i'm all in.
RAM, SSD storage, bandwidth, IO are still shared.
It's pretty much a VPS, not dedicated.
RAM and CPU are NOT shared.
How does the RAM sharing work? Are there any public details on this?
There is no RAM sharing, it's a metal server
I thought that might be the case :-)
The bandwidth is capped far below the max I assume, does that not provide a stable 200 mbit for everyone?
Unlikely...If there are 912 blades per server it would require 20Gbps of dedicated capacity per server. Having played around with Scaleway servers, they have at least a two cabinets of these servers (10-20 servers).
Online.net's total network capacity is in the realm of 1Tbps.
They would have to dedicated anywhere from 20-40% of their network capacity if they were to guarantee 200mbit
of course we don't deploy a network accounting for every machine at 200mb/s flat 24/24, this would just be throwing money away. The reality is that most users won't be using these 200mb/s constantly, every customer have different traffic pattern usages. We (online.net) have a long history in dedicated servers (yes we have a >1tb/s global network footprint), we known how to scale the network the proper way. It is carefully monitored, when upgrades are needed, we do upgrade (for example you can check on status.online.net network section, you will see that we upgrade when it is needed, we have quite a few ongoing) the same applies to scaleway, we can upgrade network uplinks if needed and we do monitor that closely Note that only uplinks to the backbone are "shared", each 1gb/s port of the servers really are dedicated Hope this clarifies the design :)
Do you mean the 1Gbps port for the enclosure (i.e. the server hosting the 912 blades)? If so, unless you're providing many 1Gbps links you're not anywhere close to providing good quality networking for these
every internal links (between blades, chassis and backbone) are x*10gb/s :) server to switch is 1gb/s dedicated
Comments on HN normally provide both sides to the story, so I am hoping someone with deep/vast server knowledge and the VPS market will provide a critique of why this is "not so good".
Also, France? Same country that is trying to push privacy-unfriendly laws? Probably not a good idea to use this service for VPN then.
It would probably suffice as a test-box or cheap-static hosting I guess.
> Also, France? Same country that is trying to push privacy-unfriendly laws? Probably not a good idea to use this service for VPN then.
They are explicitly advertising VPNs, but even though I live in France I'm not sure whether the situation there is secure enough. France is collecting internet logs, and a server has its own IP, is a VPN then even useful?
> It would probably suffice as a test-box or cheap-static hosting I guess.
More than static. I installed serendipity (a blogging engine, think wordpress) there and it worked good enough for me to assume that it would work fine with medium traffic as well.
Got a pretty good impression otherwise. The thing that bothered me most was that the free test month was not a full month, but till the next accounting period – 14 days for me. Cut my time short to test it. But you can imagine that if that is my sole complain, it was fine otherwise for me. A solid site and their scripts to build images worked fine.
If your concerned with privacy friendly VPS then you really can't purchase virtual servers in any western country.
I suggest you look into eastern europe where you can purchase server time in BTC.
Boxes like these are extremely popular as BitTorrent seedboxes. Online.net, for example, is huge, and not just in Europe; people run FTP servers on these things and copy stuff over to their home servers. I would say that $3.40/mo is pretty competitive.
> Also, France? Same country that is trying to push privacy-unfriendly laws? Probably not a good idea to use this service for VPN then.
As opposed to all those famously Switzerland-based companies such as AWS, Heroku, DigitalOcean...
Probably not the best place to run a seedbox either.