"true cloud" providers

Ree

New Member
So say I'm lazy (and I am!) and want high availability without having to do it myself.  Seems like "the cloud" is my answer, but not all clouds are created equal so I'm having a tough time identifying providers.

So who are the "true cloud" providers, which to me means at the very least they have automatic (not manual) failover of VMs to another server in case theirs goes offline.

And does anybody use a fully redundant SAN to eliminate that as a single point of failure?

How about data center redundancy, so if power is cut and generators sieze at datacenter A, then everything spins back up in datacenter B?  (I know the costs would be insane for this, so this is more of a curiosity thing than anything)
 

datarealm

New Member
Verified Provider
We are offering cloud servers on cloudstack with high availability enabled.  So if your VM fails on one node it will automatically be brought up on a new node.  At this time we use a single file server per cluster, but within that server have taken measures against failure:  raid 1 SSD cache drives, raid storage array, multiple power supplies fed from different PDUs.

I'm not sure what other cloud providers offer this type of automated failover of VMs.  I am pretty sure rackspace and AWS do not.  On the other hand you can still architect highly available sites on other systems using multiple VMs and either implementing IP failover or load balancers...
 

serverian

Well-Known Member
Verified Provider
How about data center redundancy, so if power is cut and generators sieze at datacenter A, then everything spins back up in datacenter B?  (I know the costs would be insane for this, so this is more of a curiosity thing than anything)
I doubt if anyone in the industry does this.
 

tchen

New Member
I doubt if anyone in the industry does this.
There's vmware which supports stretched clusters (~100km apart at most).  Although anyone providing this as a end-client solution probably has them closer together to ensure the storage backend can do synchronous replication without losing too much performance. 

Frankly, to the OP - don't rely on IaaS HA solutions.  For most people, making sure a good backup exists (and possibly master/slave database) along with a clear disaster recovery plan go a long way to ensure uptime.  That's a robust plan, and you're not beholden to 'fully redundant SAN' downtime.  :D
 
Last edited by a moderator:

Dylan

Active Member
Prometeus' iwStack is probably hands-down the best HA solution you're going to find here -- CloudStack, KVM, on a heavy-duty SAN, at an awesome price (under 4 euros/mo for a 512mb instance with 15gb of storage). 

http://iwstack.com/
 
Last edited by a moderator:

Ree

New Member
datarealm: I saw your ads on WHT when I was searching earlier.  Searching for your company name just brings up your ads without much discussion threads -- are you still new and not used by many clients yet?

serverian: Yeah I don't need multi datacenter redundancy, was just curious since I was already asking the related question.

tchen: I totally agree, for me all I have is a script I can follow to bring up an identical server, and regular backups of /var/www.  I'm asking for a research group that is overpaying for an underperforming service from some local web firm though, and so if I go to them with a recommendation I'll be a little pickier than usual.

Dylan: Very cool.  Something I might even consider for myself if/when they come to the US.
 

tchen

New Member
If the datacenter resilancy is off the table, then any OnApp provider should be able to float your boat. I wouldn't call it a recommendation though as they tend to charge far too much for snake oil.
 

MannDude

Just a dude
vpsBoard Founder
Moderator
If the datacenter resilancy is off the table, then any OnApp provider should be able to float your boat. I wouldn't call it a recommendation though as they tend to charge far too much for snake oil.
I'd be weary of just any ol' OnApp provider. They were doing that free license thing for a while, at least for a certain number of cores (I believe). Lot of folks with inadequate setups jumped the gun because 'omg cloud1!!!' and well, it can't be compared to someone who invested properly into their service.

Good thread though. Curious to see what other recommendations pour in.
 

VPSCorey

New Member
Verified Provider
We can do this with 2 vps on 2 nodes with a load balancer front ending the apps.  We even support fencing to keep them from firing up on the same node if node A fails and the 2nd vps is on B, it reboots the failed VPS on any other node except B.
 

shovenose

New Member
Verified Provider
Used Hivelocity's SparkNode Cloud a while back. Most horrible experience ever. I know that's more of an unrecommendation, but just keep in mind to avoid them.
 

Enterprisevpssolutions

Article Submitter
Verified Provider
Onapp is buggy. We have HA for our cloud systems redundancy is the key but its also on how you restore when everything fails that you have to worry about. Clients are always told to keep backups as you can never be too safe.
 

willie

Active Member
I don't think of automatic failover as a characteristic of cloud providers.  It could only help with total failure of the VM anyway, which doesn't happen that often.  You need your own monitoring and failover for everything else.

Characteristics I see as important:

1) Hourly billing: if you have a sudden burst of demand, you can add more servers for a few hours or days, and not get charged for a whole month.

2) Constant availability of more servers in actual practice, never out of stock.  That means the host needs enough hardware to meet the highest level of demand, e.g. during holiday shopping season.  That in turn means the host either has a lot of idle hardware the rest of the time, or has something like Amazon Spot instances which are sold at lower prices but can be preempted by higher paying assignments (On-demand instances).

3) Storage separate from VM's, like S3 and EC2

4) availability of high-utilization instances which basically means single-tenant servers (you're still on a VM, but it's a dedicated server underneath, you're not sharing the hardware with anyone else and you can use all the machine resources)

Technically not a cloud characteristic but important IMHO:

5) availability of large machines: the biggest EC2 instance has 256GB of ram and 16 cpu cores, unless they've added something even bigger by now.  There are times when I've needed something like this for a short period, for some particularly large computation.
 
Last edited by a moderator:

datarealm

New Member
Verified Provider
datarealm: I saw your ads on WHT when I was searching earlier.  Searching for your company name just brings up your ads without much discussion threads -- are you still new and not used by many clients yet?
Heh, no, we just don't do much advertising. :)

Datarealm has been hosting since 1995.....
 
Top