John,

You’ve raised some great questions about SDC/Triton that I’d love to try to 
clear up inline.

> On Jan 12, 2016, at 8:07 AM, John Barfield <[email protected]> wrote:
> 
> The funny thing is that, I hadn’t realized that Joyent was charging a premium 
> for the self-service portal.

The portal and its licensing are an accident of history and circumstance, not 
the long-term intention of Joyent, and it’s not intended as a “gotcha” to force 
people to pay. What Joyent really charges for is support, and only for users 
who think that support would add value.

> Another thing that I’m not fond of with SDC is that the controller requires a 
> dedicated server, which can be a single point of failure.

This is a complex question. The failure of one or more SDC/Triton services, 
including the entire head node, does not cause the data center it manages to 
fail. Provisioning or other services likely won't work during a failure, but 
any running “customer” instances on compute nodes will continue to operate 
without interruption.

Persistent data used by the SDC/Triton services is distributed across multiple 
nodes, so the total loss of any single node (or multiple nodes) will not cause 
the loss of that data, and the data center can be recovered.

As gentle as the failure modes I just described are, not everybody is 
satisfied, and we recognize there’s room for improvement. As should be said 
about every open source project, pull requests are welcome.

> (I do *think* that SDC is multi datacenter aware or Multi-SDC cluster aware, 
> which could be why this design was chosen, maybe someone at Joyent could 
> chime in on this. Back when I designed AppLogic grid implementations I always 
> told folks to setup multiple grids and not to just rely on one. Then 
> architect your apps such that it fell over between clusters in the event that 
> 1 entire cluster was down).

Yes, distributing an application across multiple independent data centers is a 
very, very good idea and something we recommend to all. (Though Joyent was just 
named the most reliable cloud provider of 2015: 
http://www.networkworld.com/article/3020235/cloud-computing/and-the-cloud-provider-with-the-best-uptime-in-2015-is.html
 .)

> This is actually one of the reasons why I chose Fifo to begin with. The 
> controller can be clustered on as many machines as you’d like and it runs in 
> zones instead of on bare metal. Each node is independent.

To be clear, SDC/Triton services run in zones, not on bare metal, and as 
described above, each compute node and any “customer” zones on them continue to 
operate even if SDC/Triton services are unavailable.

- - - 

Please understand that I’m only replying to describe SDC/Triton’s features and 
behavior, not a comparison to Fifo, and I offer no judgement to either’s 
applicability to your needs.

—Casey


-------------------------------------------
smartos-discuss
Archives: https://www.listbox.com/member/archive/184463/=now
RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb
Powered by Listbox: http://www.listbox.com

Reply via email to