DUKE ITAC - October 10, 2002 Minutes

Minutes

October 10, 2002

Attending: Mike Pickett, Pat Halpin, Mike Baptiste, Melissa Mills, Rafael Rodriguez, Billy Herndon, Chris Cramer, George Oberlander, Ed Anapol, John York, Mark Chuman, Kevin Smith, Clare Tufts, Robert Wolpert, John Board, Bill Rankin, Steve Woody, Scott Lindroth, Bob Currier, Rob Carter, Roger Loyd, Molly Tamarkin, Fred Westbrook, Tracy Futhey, Lynne O'Brien, Craig Henriquez, John Harer, David Jarmul, David Kass for Brian Eder

Guests:

Call to Order: Meeting called to order 4:05 pm

I. Review of Minutes and Announcements:

Announcement of talk on student center learning. Lynne O'Brien.
Wednesday, October 16, 2002 - Student Centered Learning in the New Millennium by Dr. David G. Brown

II . Discussion: Supporting Beowulf Cluster Computing - John Harer

A. Issues

  1. Support --Parallel Coding --Installation --System Administration
  2. Space
  3. Power
  4. Networking
  5. Software
  6. Storage
  7. Utilization
  8. Costs
  9. BioGrid

- 32 clusters exist in variety of locations - Owned by variety of research groups and departments - Primarily in sciences and engineering - Supporting clusters is a challenge - Installation and maintenance of clusters must be done - Bill Rankin was hired - he helps facilitate using/support for the clusters for people around campus - The clusters take a lot of space and power and cooling - The rooms need to be set up a certain way to support these, there are software challenges, etc. - It is not clear if machines are being fully utilized - A lot of money is spend for high-end capability and not sure how much is being used - Bio Grid - clusters first phase

B. Central Support Models

  1. Do nothing - see what happens
  2. House Clusters with Researchers - Provide Shared Support (go around and provide support, does not address space or power, etc)
  3. Central Facility and Support - "Rent a cage" (here is your cluster/space - doesn't deal with utilizing cluster)
  4. Central Facility and Support - "Cluster Farm" - charging model

Melissa Mills - Can you do combination of "rent a cage" and a "cluster farm"? Have an option 5, do either one? Different people would have different needs. Offer both options?

John Board - Many groups bought clusters to get away from sharing resources. A combination of all options might be best.

David Jamieson Drake - will this be a commodity service on Internet 2? Are we going about it the right way to lead up to it being on Internet 2?

John Harer: We can scale things up and down - we have a national lab to serve us

Chris Cramer: When you are talking commodity, it is the hardware, lifecycle is only 3 years, investment is really in the machine room space, could always use it

Melissa Mills- Want cycles and buying clusters and don't know where to put them

Mike Baptiste - Have Duke recommended configurations, nodes, etc.

Molly Tamarkin - it would be worthwhile to get faculty together that would use this and see what their needs are

John Harer - they are already getting together people who have them

Tracy Futhey - could also have a group of people for those using clusters and those that want to use them

Mike Baptiste- Need to think about the infrastructure issues like cooling, etc

Fred Westbrook- Think it is important to push a hybrid solution, instead of centralized use if institutional funding

C. Costs

  1. Up Front --Renovation --Equipment --Networking
  2. Ongoing --Equipment --Systems Administration --M&0

III. Observations from 2002 Educause meeting - Melissa Mills, Molly Tamarkin, Lynne O'Brien, Billy Herndon, Tracy Futhey

Lynne was focusing on instructional applications- -There didn't seem to be a lot of anything new - hardware, software or activities -Seems like everyone has been busy upgrading administration systems and are now they are just beginning to think about education. -Many sessions - faculty related to using digital media -There was a talk about digital collections, how to make them, index them, use effectively -Uses of handheld computers in classroom and emerging software for it -Course management systems - debate - home grown vs commercial

Melissa - webpage See : http://www.aas.duke.edu/comp/educause/2002.html -Security awareness - there was a lot of interest in security awareness- UVA - open knowledge - talking about open source, interested in letting Duke join the meeting at NC State and getting together with Duke while they were there -Project management means of sharing info between it people and managers where resources need to be put

Molly Tamarkin- -organizing, managing, funding -all lessons learned - process they used versus what they were doing -managing people is WAY harder than managing computers -project management -middleware - open source middleware

Billy Herndon- -Ivy plus counterparts, meeting in October - Yale is hosting on total ownership on ERP -Web development, centralized web services, public relations - policies and procedures, central resources, application development, teaching and learning policy regarding web - logo, photos, templates -Best practices for student services - Univ of Delaware - a cross functional one stop shopping, Boston College, changed facilities -Perdue - epay system - gone to pushing more of payment of checks as opposed to credit cards , parents needed access to student info..discussion on FERPA -eportfolio application - digital collection of files and forms students might pull together.

Tracy- Spoke to Sun, Apple advisory committee, Ivy plus, etc. Common theme: community collaboration, how do we get together and do things ourselves getting together on uportals, digital libraries, how do we do these things together. more sense of working on things together -One session - Drexel spoke of their CMS implementation - would be good to get the tape of that and get together with content management systems team Drexel learned a lot of things in their process. It is good to be aware of these things before we get to that point.

IV. NetReg - Demo of proposed ResNet PC registration process - Chris Cramer, Bob Currier

How do we get in touch with the student whose computer seems to be compromised? Klez tells you what the machine is. They would like to get in touch fairly quickly with students who have viruses or are hacked. Now, they pull the IP address, ask Datacom to find where the IP address is and they go out to the dorms, etc, within 2 hours - 2 days, they can tell what room number has the IP address and can get with housing, to find the occupants. It would be nice to locate owner of computer faster

Proposal: -When the student brings up computer on the net for the first time - he or she has to register their mac address in their name on a webpage -Once they have their NetID, they can contact them, contact quickly -Find out from DHCP logs what mac address is registered to which students -Low overhead from students, do once a semester register and then it would be transparent

Pat Halpin - Is the initial response authenticated?

Chris Cramer - Yes

Robert Wolpert - Why would you do this more than once?

Chris - might make sense to flush out, may get a new ethernet card, etc.

John Board - If they have a Nat, then go to a classroom, etc, will they register 3 times?

Chris- Don't think so, but need to verify that

DEMO goes to registration page enter NetID and password need to restart computer

Melissa Mills: It is good to have registration process and cleaning out makes sense It looks like a good place to start
Could possible have further discussion on presenting a tutorial or test that presents material to the students there before they can accept, it might be the one opportunity to educate

Robert Wolpert - Are there any privacy concerns?

Chris Cramer - Would not be doing anything we couldn't do before, just faster He has also spoken to Brandon Taylor in DSG about it

Chris Cramer - should be rolling out in spring.

V. VPN update - Bob Currier, Nhan Vo

On the Health System side they are using a Cisco VPN modem 3080 They have 1300 users Peak time is about 150 users at same time Model 3080 is just box with 256 mg RAM with Motorola power PC, supports up to 10,000 users
Software client works for all flavors for Windows, Linux, Solaris, Mac 0SX

John Board - how rapidly growing user base? Nhan - 600 to 1300 in last 45 days

George Oberlander- Are you using tunneling ? Yes Doing strictly client to remote concentrator? Do LAN to LAN and then client to concentrator

Bob Currier - University has a similar concentrator on campus, It has already been bought and they are working with the Health System on it. Rollout - will be over the next 3 months or so

David Jamieson Drake- is campus set up different in any way? Will the university be set up the same as the health system?

Bob Currier - They will have to work through all the issues once they get the equipment in

Rafael Rodriguez - split tunneling may be different on campus than at the health system

Fred Westbrook - Who should we take specific problems to? The MCIS helpdesk? Yes.

David Jamieson Drake - Is this being coordinated with the renegotiation of the ADSL contract ? Yes

VI. Update on secure email progress - Chris Cramer -

Sept 15 - insecure email was to be shut off, that didn't happen - Sent email to 12,000 folks who potentially used insecure emailers - Going to send out another batch of more personal emails - "here are machines that you access insecurely" with another reminder

VII. Update on acpub postoffice problem - Rob Carter

- 8 postoffices - one of those 8 one died - problem: the machine's raid controller ate itself (wrote garbage over all of the disks) They migrated users to postoffices queued up messages - 23-24 hours to restore - took so long because there were 4.25 million small files to restore rather than a few big ones - About 12:30 on Wednesday the restoration finished - They had the mailboxes in restored to state they were in at the beginning of the day on Monday - Took all mail queued up and delivered it, it took quite a while. In the interim - there are twice as many users on Morcock - none on Leguin. They will rebuild Leguin when the parts arrive.

- Once Leguin is back together, they will move the users back to their normal server - There will only be 5 minutes of outage for the users to switch back.

Were people on the other 7 servers affected? - Marginally, all servers had extra traffic, mail delivery was a little slower - Roughly 12,000 people would have seen slowness

What is being put in place to avoid the problem in the future? - If you lose a CPU or post office machine, that is just replacing CPU - The real nightmare is if the data goes away - Design is being reviewed for options

Robert Wolpert- Any ideas?

Rob Carter- Possibly establishing SAN, institutional resource of storage, switch model - more flexibility - New developments: synch between postoffices, you can mirror entire postoffices, not clustering but similar, postoffices interchangeable

VIII. Other Business

None
Meeting Adjourned: 5:30 PM