[Swift-devel] Planning next set of Swift features and releases

[Michael Wilde]

The problem with VMs at the moment is that most of the resources our 
users need to run on at the moment are not VM-enabled, and wont be in 
say the next 12 months or more.

Automated deployers like ADEM show signs of working well *iff* the user 
app installs nicely.

Many do, some dont. And the ADEM model offers a nice way to find the 
sites on which an app *can* install and run cleanly. So a user can start 
with those sites, and gradually work to improve their build/install 
process to handle the sites that pose problems.

Anecdotal story: Glen tried ADEM on OOPS, guided only by Zhengxiong's 
document. With little effort he was able to run on 8 sites. He sent me 
this: "Zhengxiong and I used ADEM and got oops working on up to 8 osg 
sites. Swift script is identical to the one above. (by up to, I mean 
those are the ones it installed successfully on for Engage VO, and we're 
not finished testing). "

Getting 8 OSG sites in a first test is a very promising sign - if a new 
user could do that, that would start out pretty happy.

VMs I think are definitely part of the long-term picture but not the 
near term. I dont want us to wait for VMs rather than going after the 
immediate low-hanging fruit.

On 4/22/09 12:05 PM, Mihael Hategan wrote:
> On Wed, 2009-04-22 at 16:43 +0000, Ben Clifford wrote:
>> On Wed, 22 Apr 2009, Mihael Hategan wrote:
>>
>>> I believe that the only reasonable solution to the problem is
>>> virtualization.
>> I don't think virtualisation is particularly more reasonable than the 
>> various other mechanisms. Each has its own benefits and downsides, and 
>> none of them to me appear to be particularly domininant.
>>
> 
> Well, the differential cost of going from one site to an arbitrary
> number of sites is 0 from the perspective of setting applications up. I
> think this is essential to making swift a system that commoditizes
> running applications on grids.
>