Dante is a Uniform Access Clusters similar in architecture and concept to the existing Homer cluster. The Dante system has been funded by the Student Technology Fee and will be instrumented to provide uniform access academic computing resources to the University of Washington student population.
During fall quarter we will migrate existing student accounts from Homer over to Dante. Homer will remain as the information service cluster for faculty and staff. This Homer/Dante split will allow us to architect each cluster with application and tool sets that best fit the needs of their respective populations. It will also allow us to partition the work loads to ensure scalability, availability and minimize resource contention.
The Dante user interface will be simlar to the psh interface currently in use on Homer. Psh provides an intuitive menu interface for accessing and managing system tools and applications. It eliminates the requirement that user be familiar with the UNIX command shell to make use of the system. The psh menu does support an exit to UNIX csh should a user require access to the UNIX command line.
The cluster architecture design for Dante will include:
Compute servers: Dante Application servers: Beatrice Deskmail servers: BP File servers: Guido Password servers: GuidoA remote hot spare cluster made up of one of each of the cluster server types will be included in the overall architecture for disaster recovery business resumption purposes.
Overall use policy and backup schedule will be the same for Dante as is currently in place for Homer.
"user".deskmail.washington.eduUsers will continue to use the "email@example.com" address to identify themselves to email correspondents as they have in the past.
The password space on the Deskmail servers will match that of the associated client cluster (Dante-BP, Homer-Mailer, Aagaard-EP). This is a change from the test Deskmail configuration where all the Deskmail servers share the same password space.
Quota and password management in the Deskmail environment will work like the current Homer/Mailer environment. The login "psh" interface will be used to query quotas, set passwords and manage other account settings. There will no longer be a mail shell on the new Deskmail servers as was used in the test Franklin cluster.
New user folders will reside on the Deskmail servers along with the users mail Inbox. Existing folders and user files will reside on the Dante and Homer file servers. Ftp access to the Deskmail systems will allow users to archive folders to other systems as needed.
Like the existing Homer system, Dante will use a separate cluster of system to provide outbound sendmail service.
1) Add Homer Mailers to the Deskmail service. 2) Modify Deskmail DNS to find primary Inbox site to allow multiple Deskmail accounts during migration (Example: an existing Franklin user who runs New on Dante will have two Deskmail accounts until the Franklin account is removed). There may be other ways to handle this automatically when New is run. 3) Enable New on Dante when base cluster hardware and software configuration is in place and validated. 4) Restrict New on Homer to faculty and staff. 5) Inform students that new is available on Dante and that the migration process for existing student Homer accounts will begin and continue for the next few weeks. Students whose accounts have been migrated will receive a message indicating that they can begin using Dante when they attempt to login on Homer. We estimate that we can move +1500 accounts per day. With an approximate total of 36,000 student accounts, minimum time for migration will be 24 days. Additional compute, application, deskmail and sendmail servers will be added to Dante as the migration proceeds. 6) Phase out test Franklin Deskmail service.
1) The student accounts will be extracted from the homer passwd file and added to dante with a shell of /usr/local/etc/#not_yet. This will prevent any existing accounts from being created by NEW until we can get them moved. The "#" will prevent dial-ip from finding them. 2) Movement of a student will consist of: a) The folders from ~/mail on homer/aesop will be copied to the user's dante/guido ~/mail directory. Will supply a tool to allow users to move their folders voluntarily to BP after migration is complete. b) The inbox will be copied from the user's mailer to the user's BP home directory. c) The rest of the user's files will be copied from the homer/aesop directory to the dante/guido directory. The .pinerc file may need to be fixed if it specifies local folders. Dante differs from homer in that the global pinerc file will specify remote folders rather than local folders. d) If the user has an existing deskmail account it will be merged with the dante/bp account in some yet to be determined fashion (the people with deskmail accounts will be put at the end of the list until this can be determined -- conflicting homer folders will likely be renamed "folder-homer" before the deskmail folder is copied over). e) The dante shell will be set to psh and the homer shell will be set to /usr/local/etc/#you_moved. Users will be moved only when they're not currently logged in. Only the user(s) being moved at any given time will be affected by the movement process (not counting extra overhead of the moving process).
Dates are approximate. Some activities may overlap due to resource availability. Jul 10: (COMPLETE) - Order Hardware RS/6000 E30/C20 systems Initiate disk/rack RFP Aug- Sep: (COMPLETE) - Add compute and mail servers to Homer to assist in handling fall quarter load. Assume that Dante hardware will not be in place due to vendor delivery delays and RFP process. - Validate AIX V4.2 operating system in production Homer cluster. This release will be shipped with new Dante hardware. - Install Dante response server for early queries to the domain name "dante.u.washington.edu". Oct 14-28: (COMPLETE) - Physical installation of networking, computer hardware, console connections, racks, power. - Test connectivity and basic integrity of infrastructure. - Obtain DNS definitions for hostnames. Oct 28 - Nov 1 (COMPLETE) - Complete installation of base skeleton cluster hardware. . 1 password server . 1 fileserver . 2 deskmail servers . 1 sendmail server . 1 application server . 2 cycle servers - Test skeleton cluster and create clone images when reasonably stable. - Refine and automate cloning procedures. - Train staff members on cloning. Nov 4 (COMPLETE) - Begin cloning process of existing Homer environment. - Base cluster validation. Nov 18 (COMPLETE) - Homer/Dante move message ready on student login. (DONE) - turn on Dante "new" (DONE) - Shut off student access to Homer "new" (DONE) - Begin backup and operations support. (DONE) - begin migration of students from homer (DONE) - Complete cloning process: (DONE) . 1 password server . 4 fileservers . 3 application servers . 16 deskmail servers . 32 cycle servers Dec - Migrate remaining Franklin and Mailer PC users. - Install Dante hot spare cluster at 4545 bldg.
- The tmail quota bug that Ken thought existed on AIX 4.2 has been fixed. - Large rshd memory problem on AIX causing virtual memory problems on Mailers after Pine/IMAP connection. Donn is testing Berkely and Kerberos rshd. Preliminary tests look like this may fix the problem. Will be installed on Homer and Dante clusters. (Donn) - Need to look into sendmail header rewrite of deskmail addresses. Seems to be a problem with some desktop clients that identify the particular Franklin a user is using. Ken and Steve will look at using "A" records vs "CNAME" for Deskmail servers. (Ken, Hubert) - Fix fmail so it doesn't allow "user@deskmail" address. Delivery using this address should probably work, but needs a message informing users that this is verboten. Ken recently has put some code up to address the delivery/message issue. (Donn) - Create a "deskmail.washington.edu" machine with read only Inbox to assist users who have incorrectly configured their client software. They will be instructed to use "user.deskmail.washington.edu" as their inbox server vs "deskmail.washington.edu". (Donn) - To handle redeliver problems when users are over hard quota, it was suggested that we need a tool that the user can run to temporarily increase their quota so they can bring themselves under quota. Alternative solutions included added short term checkout to the mail clusters. - Draft Homer/Dante login message (Oren) - While waiting for console connection boxes to be delivered, Eliot will use one multiport RJ box as a portable to provide console connections for cloning. The systems can run without serial console connections but they can't be installed/cloned without them.
James W DeRoest
Computing & Communications
University Computing Services