All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Need HA Scalable Cloud Instances
Hello LET,
I am looking for a scalable high availability cloud server. Not scalable as in we can upgrade you to a bigger package if you get more traffic, but scalable as in the instance will be cloned and multiplied and load will be distributed if needed. PaaS are fine, as long as they support PHP & mySQL or MariaDB (the application uses laravel, so built in support for that would be nice). I am looking for something like EC2, Google App Engine, Red Hat OpenShift etc but for a bit less than what they charge (i know its almost impossible for le* price, and im not expecting that).
DigitalOcean doesnt count since it's not truly scalable (does not have auto scaling or load distribution). The project I am working for does not want to use API's to create a scalable solution (do would work if we supported this), instead wants to use auto scaling solutions like openshift.
Feel free to recommend anything you have in mind (even do alternatives that aren't really scalable but have api, if pricing looks good i could try to work something out). Thank you!
Comments
Do iwStack.com offer what you are looking for? (I'm no expert on HA, so really don't know)
Microsoft Azure is somewhat cheaper than AWS (yes, they do Linux VMs- I have one there), though it's not 50% cheaper. They do paas too of course.
Google also does VMs - I haven't looked at them. Or rackspace for that matter.
Heroku? AppFog?
I think you are looking for PaaS instead of IaaS, and I doubt you can find anything cheaper than AWS EC2 which is IaaS btw. Auto provision, HA are value-add service, no reason the service provider will charge less than IaaS.
Good luck.
What metric will you use to determine when to provision more front end web server (processing time per request, number of non-idle workers (or active coroutines) per 5 minutes?, or more pointless load average per node) You need something to signal when to "auto scale"
Then persistence, it gets complicated, so when you scale read servers, is your write load low enough that a rw master / slave r will work, is lavarel doing something stupid like sticking sessions in the database? Perhaps an option is to optimize the stack before you consider scaling..
Most PaaS are opinionated, for example I highly doubt your setup will be compatible with the 12-factor app rules (i.e. heroku)
Hello, We can provide you HA instances with API and auto-scalable cloud storage but unfortunately we don't support yet auto-scalable performance resources. However, that's something we have on our roadmap plus with per-usage billing.
https://neximweb.com/
I think unlimited websites is similar to autoscaling?
What does that have anything to do with what im asking for? I work for them i dont make the plans. And the VM i need is for a totally different project.
Oh and unlimited websites means unlimited addon domains if you didnt know.
I'm not aware of any IAAS that does autoscaling?
Amazon does http://aws.amazon.com/autoscaling/ based on cpu usage, and there are some others out there. Most paaS like openshift do that too.
I'm trying to convince my client to use api's to do manual scaling to save resources and money in the long term, after that Ill have much more choices to choose from.
Autoscaling is often a false economy .What happens if your site / app is badly coded or so,ething hangs and your cpu or ram usage hit the ceiling? Your 'autoscaling' spins up new servers non stop for a few days and your client gets a $3000 bill? (We've seen this happen with AWS clients)
A good rule of thumb is to 'diy' and not depend on 'cloud' to give you any HA, scaling, failover or backups.
Many clouds will actually give you some or all of that list, but assuming they dont' and planning yourself will save you time, tears and money.
I agree with you. Things could go out of hand with autoscaling.
Right now Im building a prototype scaling system that creates new droplets on digitalocean based on pre built images according on cpu load (the app im planning to deploy will use a lot of cpu). Hopefully Ill be able to use it on the actual system.
BTW does anyone know of other providers that have api's and allows snapshots of vm's like digitalocean does?
Most do (including ours) tho pricing will vary a lot by provider.
That sounds like a good solution. After all, it is basically the same as what autoscaling does, and really it should not be that much code. Most of the code you need for autoscaling you need whether you are using Amazon or not (VM template and program to connect with your cluster or whatever).
iwstack, luna node dynamic, etc.
CPU is what the app im working on will use the most. So the system should be scale according to that. The only reliable solution ive seen for that so far is Amazon Autoscale.
I have been working with the other devs to squeeze the max performance out of the app by itself already. The app is a remake of an existing app that was built on a different platform, so we have a lot of restrictions on what we can do and what we cannot. The app itself isnt big but the userbase is big and once it's deployed we are not going to have time to look back and change things on the fly.
We are still looking for other solutions. I personally try to stay away from paaS since you dont have much control on the environment, but I have to go with what the client wants.
Yeah it's not really going to be much work to get this setup. I currently have a somewhat running prototype but Ill havee to clean it up a little.
iwstack looked great at first, but it wouldnt work for us because the dc needs to be us based, and iwstack has dc is italy only as far as i could see. Can you give me a link for luna node dynamic? A google search didnt bring up anything
http://lowendtalk.com/discussion/22741/luna-node-dynamic-dynamic-virtualization-platform-kvm
There are several others, I can't think of them at the moment.
So assume only bare-bones services and roll your own everything?
That "rule of thumb" is not widely followed.
Might be just what im looking for, and prices seem great. Do you know any coupons or ongoing promotions for it?
vps.net have auto scale.
They do. But it looks like their prices are almost similar to amazon's. Ill still take a look
i signed up with lunanode, and the support is good but i have been getting problems after problem.
What problems have you been experiencing?
First i created a vm in their chicago location and got a ticket opened saying their chicago location doesnt have all the features as the other location. then i created one on the other location, and the network was slow. i tried a lot of different speed test files (cachefly and some local ones from that area) and i couldnt get the speeds to go past 10mbps (was supposed to get 100mbps or 1gbitps). the support guy told me he got good speeds from his side, but using the same vm i was not getting anything. they use burstnet's servers and i have never had good experience with them. the one in chicago is singlehop but that is limited too 100mbitsps by port speed.
and their panel is super buggy. when you create a vm it tells you it was created and takes you to home page. then it takes around 5 mins for the vm to actually work and there is no notification anywhere saying its being created.
it seems like a panel that someone scratched up using php and bootstrap (and it isnt even used properly) to call openstack's api's.
and thats not all ive been getting minor bugs here and there. and one of their location doesnt even show what ip's are assigned to your panel other than the main ip (youre supposed to get /30). and io is about 80mbps according to dd tests but i dont trust those and im not gonna complain about that unless the performance actually feels slow.
overall, it just feels like their system isnt ready for production, and looks more like a beta version. i wouldn't put anything production ready on their servers untill they get the whole mess cleaned up. it's good for development since its cheap and has decent resources for price.
Have you looked at CloudFoundry's PaaS? As for the ever so-fun autoscaling, you could look at RightScale as an application manager sitting on top of a wide array of APIs.
Yeah, this is exactly my experience too.
I feel like they should upgrade their port in the singlehop's servers and stick with singlehop. Comparing singlehop and burstnet is like comparing Linode with GreenValueHost, if they want their services to be stable in longterm, that would be their best choice. Singlehop is so much better in terms of everything (disk io, network performance) but is almost twice as expensive as burstnet.
https://run.pivotal.io/ looks good. I'm gonna get a test account and give them a shot, thanks for the suggestion
I absolutely agree. I had some awful servers deployed in production by this system, and the Singlehop nodes are so much better.
We havent really had issues with burstnet, but then again we have only been with them for 2 months. so that remains to be seen. ( if it so happens that burstnet becomes unreliable, we will certainly find another co-location provider.
With regards to the download speed you are getting from your vm, we looked into it and was not able to find the cause of this issue. Some more troubleshooting is required and hopefully we will be able to identify the issue. I tried provisioning a centos 6.5 on the same node that your vm was provisioned on, and was able to get 65MB/s download.
The disk io speed is 85MB/s because the VM disk images are stored in a distributed file system on nodes with 1gbps port. We plan to upgrade the storage node ports to 10gbps to further improve io speed.
My experience with singlehop has been good. They are the most stable datacenter I came across. But they don't offer colocation, and pricing is just way too expensive to be offering any services to the low end market.
Our custom control panel needs to be worked on, it does not integrate with openstack API fully yet but it will.
Hope this addresses some of the concerns and questions anyone may have.
This is an issue with the guest VM. Tweaking per http://lowendtalk.com/discussion/comment/228710/#Comment_228710 resolves the issue.
The control panel is a simple front-end to manage your virtual machine, we expect that most customers would generally use the API.
It should be instantly visible if you provision in Scranton location. Chicago is occasionally delayed and takes longer for provisioning since it does not lazily create the disk from template.