Hi Tamim,
A couple of alternatives:
* If you would like to reserve a lot of nodes at a single location, please use the aggregate "UtahDDC InstaGENI" as it has the most available resources. That said, I've never tried reserving 100 xen nodes at once so I can't guarantee that it will work. I know 50 nodes works as a colleague of mine did it on Friday at UtahDDC InstaGENI (it took about 17 minutes to come up).
* If you would like to reserve resources at multiple aggregates, you will want to use stitching which is pretty easy using Jacks (see [1]).
That said, I have some recommendations about how to proceed to create a large topology such as this:
(1) If you are mostly interested in 100 xen nodes and a single site will do, try using UtahDDC first. This is the simplest solution.
(2) If that doesn't work, then consider using stitching and shared VLANs to create a large multi-site topology in the following way. (A number of people have tried this technique and found it to be very successful.)
a) Test your topology and setup with a small number of nodes at a single site. I can't express how important this is and how much annoyance this step will prevent. The goal is to have an experiment that works with a few nodes and the remaining steps are just making it big and multi-site.
b) In one slice, stitch between two sites of your choice using the steps in [1]. Test connectivity with ping. (The topology should be one node at each site with a link connecting the two nodes.)
c) Share the stitched VLAN at each endpoint following the instructions in [3]. You should have two shared VLANs (one at each site).
d) In a second slice, reserve 2 or 3 nodes at one of your sites connecting it to the shared VLAN for that site. This test is to make sure your setup is correct and that the shared VLANs are working. Test connectivity between the nodes in the second and first slices. Once you are comfortable with your setup, then delete those three nodes.
e) Still in the second slice, reserve 50 nodes at one of your sites connecting it to the shared VLAN for that site. Test connectivity between the nodes in the second and first slices.
e) In a third slice, reserve another 50 nodes at the other sites connecting it to the shared VLAN for that site. Test connectivity between the nodes in the three slices.
We did a tutorial [4] which discusses how to generate RSpecs for step c and d using a tool called scaleup distributed with geni-lib. I need to add the complete instructions for a and b as well (which I've been meaning to do for awhile but the gist is listed above). I'll get those done today and send a followup with the link so you can use that if you go this route.
In the meaningtime, the video and slides on [4] explain how item (2) above works in more detail. I think the first 45 or so minutes of the video is the presentation (and discussion) from that tutorial and should explain how (2) works in more detail.
Also if you go this route, perhaps send me a drawing with your setup and I can help you make sure you have a design with will work for you.
Cheers,
Sarah
[1] Do the following to reserve a simple stitched topology using Jacks:
- In Jacks, draw your desired topology.
- Add a new site
- Drag one or more of your hosts to the new site.
- Assign a specific aggregate to each site (make sure it supports stitching, see [2])
- Hit reserve