This is a copy from U Indiana.

A Brief Discussion of the Globus 1.1.3 GIIS Architecture

The new model takes a while to get used to, but it's pretty straightforward, at least from an architectural standpoint. Implementation is another story, and it doesn't help that there's precious little documentation. I have learned a lot through trial and error - here's the rundown:

The old MDS was a central galactic server which hosted everyone's information in once place. It was nice to be able to get everyone's data from one place, but most of the time all you wanted was something like a gatekeeper contact for one of your own machines, and it was a pain to have to wait for network traffic/MDS slowness to get this data.

The new model uses what they're calling GRIS/GIIS. The GRIS server (Grid Resource Information Service) is the machine-specific server which contains information regarding the machine on which it is running. It also updates specific information to the GIIS server. It is an OpenLDAP slapd process.

The GIIS (Grid Index Information Service) server is your area-centric MDS server. It runs on only one machine and can serve as wide of an area as you want it to. It is also an OpenLDAP slapd process.

So, you have a bunch of machines, each with its own GRIS, and one of these machines also has a GIIS on it to which the GRISes are updating. If you wish to do an area-wide search you can search the GIIS. If you wish to get some information on a particular machine you can search either the GIIS or the GRIS running on that particular machine. If you search the GIIS it will query the GRIS on the machine in question for you. If you happen to know the port of the GRIS running on the machine you want, it is faster to query it directly, as many of the problems with speed in OpenLDAP seem to stem from the way referrals are implemented.

Now, in order to get things set up properly, you need to do the following:

Finally, there's a nasty little bug in one of the scripts which prevents things from working properly. I have written a fix, and it's very simple. Patching the SRC file will ensure that future installs have the right version. Patching the INSTALL ensures that future deploys from that installation have the right version. Patching the DEPLOY and restarting globus will get your GRISes talking to your GIIS.

The file is grid-info-site-backend. It lives in:


There are two simple changes, they are:

line: change to

746: if test_rdn ${search_attr} $1  (take out the = sign)
760: search_rdns_expanded $1 $2     (take out the comma)

With this fix you should have things working.

I believe that the new version of the tools, globusrun, globus-job-submit, etc, all take plain hostnames now instead of those god-awful contact strings. The new gatekeepers all run on a standard port (2119 instead of 754), and the tools understand to look there first. I don't know what happens if you are running your gatekeeper on some non standard port, or if you have more than one gatekeeper on a machine.

I hope this helps! Feel free to email me with questions - I'll try to help.