> We believe that cloud computing is an amazing technology and that everybody should be able to benefit from its advantages. That's why we decided to reduce the price of the C1
Am I the only person who is turned off by that kind of marketing copy? You lowered prices because you thought you could sell more nodes, gain economy of scale, serve as a loss-leader, your costs went down, something. You didn't lower the prices because you wanted to do something nice for everybody.
Some things don't happen for only one reason. This is a business so obviously it is profit oriented, but it doesn't rule out that the company wants to actually do something good for the society.
What's really odd is that if you take it at face value, the statement implies that they didn't care about helping people beforehand, or else the price would have been lower from the start.
I think the point is the disbelief in the cause and effect relationship as presented. I too find it a little hard to swallow that the reason the price was lowered is because they want everyone to be able to benefit from cloud computing advantages.
More likely in my estimation is that the cause and effect are reversed, and they wanted to lower prices (to gain market share, users and an advertising blurb) and that has the nice effect of opening cloud computing up to more people.
It’s a pretty common practice from Xavier Niel (founder of Illiad, Free and Online) to cut prices in half to win market shares and reduce prices from the competition. He’s done that multiple times now: Free (ISP), Free mobile (MNO), Online (dedicated servers).
It has worked every time and competitors followed by reducing their prices as well.
I think there can be a thousand valid (good) reason for them to lower the price, and it is good news for most people unless scaleway raise the price back once they gain more users. (if they do this, I wont use it again)
This is brilliant because the main cost of running gear is power draw (PDUs / electrical circuits). Having OEM/ODM blade ARM setup a-la sgi cloudrack/supermicro is the way to drive costs to the floor, in a Backblaze/Google way. Unfortunately, it's a "Dell/Walmart model" hypercommodity where such a business has to maintain massive customer subscriptions to stay cash positive and still just trickles in $.
It's an interesting space, but if I were launching a cloud IaaS/VPS, I would probably optimize for the other extreme of "Apple model" premium/full-service expensive hosting that has fantastic uptime, gear and sales/support for enterprise/startup and IT/web operations... There's some more money in that and less headaches. (The most money seems to be in the upper-middle pricepoint area.)
> I would probably optimize for the other extreme of "Apple model" premium/full-service expensive hosting that has fantastic uptime, gear and sales/support for enterprise/startup and IT/web operations... There's some more money in that and less headaches. (The most money seems to be in the upper-middle pricepoint area.)
A few weeks ago, I'd say "You mean AWS", but after this morning, and a few other incidents over the last few weeks, I don't think anyone can do fantastic uptime in a virtualized/cloud environment.
There's the rub, isn't it? Companies want to compete on quality, but if you're supposed to architect your application(s) for failure, uptime becomes a lot less important. You then only have price to compete on.
I know Twilio runs in both AWS and Rackspace, depending on various KPIs for shifting load. Seems to be the way the world is headed. I'm curious if containerization like Docker predicts CDNs running your containers for you at the edge...
Is that just uptime as reported by the OS, or actual time between outages? The server can be happily churning away even if its network connection goes down.
I've been on Linode for a while, and have also had a decent experience. However, I wouldn't use uptime as my only metric.
I will vouch for Linode. Also if you've never heard of Ramnode, they are my go-to VPS provider. I build actively monitored infrastructure with downtime notifications on the order of 5 minutes or less, so my story isn't just anecdote. I have bought hundreds of VPS over the years from a huge number of providers, and Ramnode is hands down absolutely the best provider. Excellent uptime and the VPS always feel snappy and responsive. It never feels oversubscribed like every other provider out there. Linode is also very good, too, IME.
I've had a similar experience on Linode as well. In fact, in the few months that I have been using Digital Ocean for a personal server they have had more outages than the past ~2 years on Linode.
DO has really gone downhill. For some time they were top-of-the-heap for me with Ramnode and Linode, but lately (the last two years or so), they have really gone downhill with respect to overall quality.
For every comparison we should take into account that scaleway offers dedicated hardware, not a VPS.
Also I think its important to note that they (currently) only offer one very "small" server model, so your whole application would have to be able to scale horizontally really well to be able to run on such infrastructure. So you can't have a few big database servers and a lot of small stateless application servers, which I belive is a very typical architecture today.
But... Scaleway is a spinoff of Online.net which offers fine (and cheap) dedicated Intel servers. I don't know (yet) the ping between Scaleway and online.net but it's probably low enough to allow scaling "workers" on ARM servers, while DBs are kept on 64bits dedicated Intel servers.
Also if you want to scale horizontally, you probably want to scale on more than one datacenter (online provides that too).
I have a running server at online.net (small personal tier, 8 core, 8gb RAM, 120gb ssd) [1] for 20euros/m running a CI server. I've spinned up a scaleway server and ping from online.net to scaleway or reverse is under 1ms (0.5-0.8ms)
Also, they are offering VPC by default, the nodes seem to have only one network interface which defaults to private network, and external traffic is routed to private network with specific firewall rules. Also, they are offering ip failover, I think you can attach or move external ips at runtime between servers without configuring anything on the nodes (just from API).
For online.net I know they are offering DDOS protection for higher priced servers, I'm now sure if they are supporting this for scaleway, but you can always reverse proxy from online.net server to scaleway to achieve horizontal scalability if this is an issue.
Regarding DDoS protection, if you scroll down on the Scaleway pricing page, there's an addon "Premium bandwidth with full DDoS protection" for €0.02/GB.
You are correct that you maybe can combine that with dedicated servers from online.net, but managing that might be a headache. Scaleway presents itself as a cloud service after all, where easy/automated management, quick scaling, rapid deployment etc is an important part of the equation.
Good point that they seem to be missing multiple availability zones.
It looks pretty clean to me to have dedicated intel servers with lots of memory and local disks for DBs, and workers that can be easily started with an API!
We are currently running on online's dedicated intel servers (among others), but we've kept an eye on their ARM offering...
speaking of online.net, they currently have a banner ad front and center on their homepage for scaleway which says:
"Run up to 10 bare metal SSD cloud server for free" [1]
Clicking through just lands me on the scaleway homepage. Can anybody shed some light on this? Clicking around their site and couldn't see any mention of it.
Scaleway offer the first month for free, and the maximum amount of servers you can have running during the free trial is limited to 10, to benefit just signup on https://www.scaleway.com
Scaleway servers are hosted in Online.net's DC2, you can get dedicated servers in the same DC, or in another one of their DCs which are all directly linked to one another.
Storage and VPS systems are becoming (have become) commodity so quickly.
It will be interesting to see if this has any effect on open-source Saas businesses. I.e. "We'll host it for you or you can host it yourself for free" - for example, ghost.
Hosting it for me has definitely has its level of remove-hassle advantages, but if such software is easy to install and low maint, then it becomes notably cheaper to grab a VPS for it.
Well, having just now tried this out, I have to say I'm happy with their interface and the experience getting a server started. I'm excited to see what the uptime is like on this new box and if the unmetered bandwidth is truly unmetered bandwidth.
C1 is probably not an extremely performant machine, so what is the preferred solution for database hosting? I guess the latency will be too high if we host the database remotely i.e. outside Scaleway but the C1 instance might be not good enough to host a database which is normally the main bottleneck?
This makes them very similar in price to Digital Ocean, even slightly cheaper. I wonder what the performance comparison would be between their equivalent tiers...
Their S3-compatible object storage is also quite appealing.
Being in North America though, I'd love if they had a data centre option here too.
It's 180ms from a residential Canadian connection so I'd imagine it's similar from some parts of the US. Expecting ~300ms from my Australian ADSL connection (compared to ~190ms to an LA based DigitalOcean instance).
vpsbench on Scaleway:
CPU model: ARMv7 Processor rev 2 (v7l)
Number of cores: 4
CPU frequency: MHz
Total amount of RAM: 2022 MB
Total amount of swap: 0 MB
System uptime: 10 min,
I/O speed: 99.6 MB/s
Bzip 25MB: 32.62s
Download 100MB file: 70.5MB/s
Small DO instance:
CPU model: Intel(R) Xeon(R) CPU E5-2630 0 @ 2.30GHz
Number of cores: 1
CPU frequency: 2299.998 MHz
Total amount of RAM: 494 MB
Total amount of swap: 1023 MB
System uptime: 40 days, 11:26,
I/O speed: 445 MB/s
Bzip 25MB: 8.76s
Download 100MB file: 25.8MB/s
I just created a test instance with them and ran an open source project of mine. It runs just as fast as anywhere else, but compiling it takes 5 times as long. Investigating further to see if it's just something about my (weird) project.
Edit: ah pella posted a comment at the same time as me with details on just how slow the ARM CPU is: https://news.ycombinator.com/item?id=10161307. The benchmarks linked there are 15x slower on single-threaded workloads.
Amon's piece is back when Scaleway was way more expensive. Amon says that he started a DigitalOcean Droplet with similar specs to the Scaleway box.
The Scaleway box has 2GB of memeory and 4 cores.
The first Digital Ocean box with more than 1 core and 2GB of memory is $20 a month. That's 6 times more expensive than Scaleway!
It would be interesting to see what the benchmarks would be like on the price competitive Digital Ocean offering (single core, 512MB memory, 20GB Hard disk)
Scaleway, using one of four 1332 bogomips processors and 2GB RAM (~$3.30/mo):
Test execution summary:
total time: 685.4730s
total number of events: 10000
total time taken by event execution: 685.4637
per-request statistics:
min: 68.52ms
avg: 68.55ms
max: 75.46ms
approx. 95 percentile: 68.58ms
Digital Ocean, using the sole 4600 bogomips processor and 512MB RAM ($5/mo):
Test execution summary:
total time: 34.9343s
total number of events: 10000
total time taken by event execution: 34.9257
per-request statistics:
min: 2.88ms
avg: 3.49ms
max: 8.00ms
approx. 95 percentile: 3.68ms
benchmark : "Scaleway vs DigitalOcean and deploying web apps on an ARM server" ( By Martin Rusev on 28.05.2015 )
"In all my tests, the performance of their servers was slower compared to DigitalOcean, but this could be because they are using Network drivers(LSSD) and a 32bit ARM architecture(armv7l)"
SSD drives provide excellent performances with a really low latency. They perform especially well with random access patterns. Our SSD storage provide an average latency of 550 μs (550 microseconds) when a traditional rotating drive has a latency of several miliseconds.
We provide 2000 IOPS (an IO is a 4K random read or write access) per 50GB chunk you buy. It means that if you buy a 150GB volume (3 chunks), you will get about 6000 IOPS (3x2000).
To enjoy the performance of your volume, your application needs to use parallel IO. The total bandwidth of your volumes is limited to 120MB/s by the network link."
They seem to provide much more memory at a slightly cheaper price, but given that it is arm I am not sure I care - can you run Ruby, Python and (Oracle) JVM?
I really just wanted a cheap server for Minecraft.
Scaleway is a highly interesting player in the IaaS market as they're one of the few currently that are offering ARM based servers. Will we see more ARM servers the next couple of years from more vendors?
Total isolation is also interesting for some high-security applications.
I Tweeted to them and suggested that they offer -- at a possibly much higher price point -- a physically isolated server that must be manually provisioned by a human being. Any access would require multi-factor authentication followed by the actual dispatch of a person to retrieve the device.
This would be interesting for things like CA master signing servers, etc.
Almost certainly. There is another generation of ARM SoC's coming out with more and beefier cores, specifically designed for servers. One thing I'm curious about is whether this price reduction from Scaleway means they are close to announcing availability of those.
I wonder if it is effective to run Tor nodes on this, given the unmetered bandwidth. I suspect that CPU is going to be a huge bottleneck here. Does anyone mind trying it out?
Don't bother trying to run an exit relay. The operators of the company will assume you're a terrible person and forward your information to law enforcement. [0]
That's very interesting. I wonder how those ARMs can deal with encryption (VPN). Also OpenBSD image or, better, ability to install anything from ISO would be very nice.
Not yet, this is in our mid term roadmap. We developed our own routers and switches to get as many features as possible. Some work is still needed to integrate IPv6 with all security features."
(The title currently just says '2,99 per month'. This should probably be changed "2.99€" instead, since a lot of people would expect HN to default to USD. Even I (living in Germany) was surprised it was in Euro.)
I didn't check if they do, but they should not charge VAT for customers located in the USA. Which means that the price will actually end up being $2.80 (as of right now 2.99/1.2 EUR to USD).
I can think of several good reasons to use American Express, despite it being somewhat less widely accepted. But Discover doesn't seem to have any obvious value over a MasterCard or Visa.
Discover support has been top notch in my experience and after having a number of issues with other credit card providers they've been my goto for a number of years now.
What about American Express? They're doing pretty well. I don't think I've ever seen anyone use Discover, so I don't know how much competition they really provide.
Afaik they're mostly accepted in Europe via their partnership with the payment processing network of Diner's Club (vs. directly seeing a "Discover" sign). Diner's Club is moderately widespread but only in certain market segments, like restaurants and higher-end stores. Not many European ecommerce shops will take Discover or Diner's Club, not even big ones like Amazon [1].
"Today, we are announcing a huge price cut for Scaleway compute services. We believe that cloud computing is an amazing technology and that everybody should be able to benefit from its advantages. That's why we decided to reduce the price of the C1 to €2.99 per month or €0.006 per hour, effectively reducing the price by 70% compared to the previous price (€9.99 per month)."
Am I the only person who is turned off by that kind of marketing copy? You lowered prices because you thought you could sell more nodes, gain economy of scale, serve as a loss-leader, your costs went down, something. You didn't lower the prices because you wanted to do something nice for everybody.