[kepler-dev] Metacat and MyProxy Running on Roadrunner (fwd)
Jing Tao
tao at mercury.nceas.ucsb.edu
Thu Aug 18 12:07:11 PDT 2005
Jing Tao
National Center for Ecological
Analysis and Synthesis (NCEAS)
735 State St. Suite 204
Santa Barbara, CA 93101
---------- Forwarded message ----------
Date: Thu, 18 Aug 2005 13:48:55 -0500
From: Bill Baker <bbaker at ncsa.uiuc.edu>
To: ltergrid at nsf-middleware.org
Subject: Re: Metacat and MyProxy Running on Roadrunner
To follow up on the questions raised in yesterday's conference call,
1. Incompatibility between GSI server-side connector and Tomcat 5.5:
Yes, what Terry said. Some of the Tomcat classes that the connector depends on
have been repackaged in 5.5, and so the connector can't link to them. I
suspect that it wouldn't be hard to update the connector, but I haven't tried.
2. Duplicate LDAP UIDs:
We configure OpenLDAP, which provides the PAM back end, to only look at one
organizational unit, specifically "o=lter,dc=ecoinformatics,dc=org", within
LTER's LDAP. Within each organizational unit, the UIDs are unique.
Practically speaking, I don't know how we would manage multiple LDAP domains --
theoretically, I can think of a few alternatives, but they all have drawbacks.
- Extend the PAM sequence to include organization in addition to UID
Problem: MyProxy currently only supports username+password
- Encode the organization name into the PAM username
Problem: that's a lot for people to remember
At 09:21 2005-08-18, Von Welch wrote:
> Bill,
>
> Thanks for the update. Questions and comment below.
>
> On Aug 17, 2005, at 5:14 PM, Bill Baker wrote:
>> - running under Tomcat 5.0.28 (Tomcat 5.5 is not compatible with
>> our
>> GSI server-side connector)
>
> What are the details of this incompatibility?
>
>> MyProxy:
>> - configured for LDAP PAM for password authentication
>
> Following up on the conversation from yesterday, what happens today
> if we use a uid that is duplicated in LDAP?
>
>> Questions:
>> - Is there interest in running a CA on roadrunner, so that MyProxy
>> can originate its own certificates? They would require (simple)
>> command-line initialization for each new user, but users wouldn't
>> need NCSA credentials. On the other hand, only grid resources
>> that are configured to trust that CA will accept them.
>> Such a CA could peacefully coexist with the current setup.
>
> I don't see strong benefit for this in the scope of the demo.
>
> Von
>
>> -- Bill Baker
>> Visiting Research Programmer, NCSA
>>
More information about the Kepler-dev
mailing list