August 2001 Email Thread
X-Sender: keahey@localhost
Date: Thu, 30 Aug 2001 16:10:26 -0500
To: "Thomas W. Fredian"
From: Kate Keahey
Subject: Re: MDSplus with globus_io
Cc: Mary Thompson , Martin Greenwald ,
"Schissel, Dave" ,
"McCune, Doug" , Ian Foster
Status:
This is wonderful news! I will try to address the rest of your questions
tomorrow at the latest; have to run now. Also, I got your account
request for ANL and it is in the process of being authorized.
At 04:48 PM 8/30/2001 -0400, Thomas W. Fredian wrote:
Hi Kate,
Mary told me of a linux system at lbl I could use. I was able to debug my
problem and I now have the MDSplus mdsip service running using globus
authentication and gsi_wrapping of the packets! I was able to communicate
successfully between a linux system and a sunos system. I used tcpdump on the
packets to convince myself that they were indeed getting encrypted. There is a bit
more work to do to have this part ready for the demo but it seems to be working
fairly well now. Right now the server just prints out the identity of the
credential used to connect to it. It does not validate it or map to a user. This
should be easy to do but shouldn't be critical for the demo. I still will need to
know how one makes a globus service which gets run by inetd (each client gets his
own server process when it issues the connect call). The only other question I had
was whether or not you can alter the handle attributes of a globus connection once
it has been established. I am hoping to gsi_wrap only the mdsplus header portion of
the client server communication and not the whole data transmission. This would
entail switching back and forth between gsi_wrap and clear on the same handle. I
haven't tried this to see if there is a problem doing so.
The problem I was having was caused by code being executed in an unexpected
order. In my server code which accepts connections from clients I was registering a
read callback and then setting up some client data structures. The read callback
was getting executed before my accept code completed. This sure looks like threads
even though I linked to the nothreads globus build libraries. Is there are globus
standard way of creating mutex locks so I can make this thread safe?
Thanks,
Tom
Kate Keahey wrote:
> Tom,
>
> This is wonderful news about MDSplus! We have to be careful that you don't
> solve all of the Fusion Grid problems by SC ;-).
>
> Bug: let's try looking at it under a debugger. If you could request an
> Argonne account I will try to get you the exact software configuration you
> need. f we look under the debugger and nothing happens (nothing simple
> anyways) then the thing to do is to go to the problem reporting page as I
> described in an earlier mail. Send me the number as soon as you get it and
> I'll make sure things get fixed.
>
> For an account: go to https://www-accounts.mcs.anl.gov/ and fill out the
> application form naming me as a sponsor. My signature should contain most
> of the info you'd need (but let me know if that is not enough). This is a
> new thing, we'll see how well it works. It might ask you for resources, say
> "DSL machines". Let me know if the process is annoying.
>
> More below.
>
> At 04:19 PM 8/29/2001 -0400, you wrote:
> >Hi Kate,
> >
> > The examples did help. The MDSplus client is now successfully passing
> >credentials to the server. I'm trying to track down a problem I run into
> >when I
> >turn on GSI_WRAP. With clear mode it all works great. If I turn on
> >GSI_WRAP a few
> >mdsplus messages get sent back and forth successfully but then the server
> >segfaults
> >somewhere. I would love to have an account at Argonne also (preferably on
> >a linux
> >system!). I can't seem to get gdb to work correctly on portnoyc (SunOS) so
> >I've had
> >to debug this thing using brute force (printf's everywhere!).
> >
> > I can now get the identity of the client connecting to the server so
> > I could
> >use this to look up a mapping to a local user using the same code I had in the
> >server for validating user@host. If there is already a globus capability
> >to do this
> >(i.e. given a credential find out if the client is permitted to use a
> >service and
> >which local account that service shoud run under) could you point me to an
> >example?
>
> There is a capability like that: every machine that has Globus installed on
> it contains what is called a gridmap file that maps your credential to a
> local id on that machine. But I do not know if it is accessible through the
> external API. Let me look.
>
> >-tom
> >
> >Kate Keahey wrote:
> >
> > > Tom,
> > >
> > > Here is a tarzipped example; this was the best way for me to learn about
> > > the usage and it should also answer your questions. But please keep the
> > > questions coming if not. Both client and server need to call the functions
> > > to be compatible. In order to obtain credentials you call globus-proxy-init
> > > and that takes credentials from your .globus directory and generates
> > > temporary credentials to use in your session with globus. Would you like to
> > > get an account at Argonne also? It might be easier for me to point you at
> > > examples and reproduce problems if any if you do. Let me know.
> > >
> > > At 04:38 PM 8/28/2001 -0400, you wrote:
> > > >Hi Kate,
> > > >
> > > > I found those routines but there isn't any explanation on how one
> > > > uses them.
> > > >Having no experience dealing with secure communications it isn't very
> > > >clear who
> > > >needs to call these (client or server), when and what to use as arguments. For
> > > >example, who calls globus_io_attr_set_secure_authentication_mode? and what
> > > >do I
> > > >use for the credential argument? I'm not sure where to find a globus_io
> > > >example.
> > > >I'm working with this on a sun system at lbl.
> > > >
> > > >-tom
> > > >
> > > >Kate Keahey wrote:
> > > >
> > > > > OK, let me look into your question.
> > > > >
> > > > > For Globus io documentation look at the attribute functions. You can search
> > > > > on "set_secure_authentication_mode", and there they all are more or less.
> > > > > Also channel wrapping. And let me know if you encounter any problems at
> > > > all.
> > > > >
> > > > > At 04:03 PM 8/28/2001 -0400, you wrote:
> > > > > >I'll look some more at the globus_io documentation.
> > > > globus_io_tcp_connect is
> > > > > >used by the client to connect to a server. What I need is some
> > > > mechanism for
> > > > > >the server to begin using globus_io on stdin (which is the socket
> > > > connection
> > > > > >to the client when the application is run by inetd).
> > > > > >
> > > > > >-tom
> > > > > >
> > > > > >Kate Keahey wrote:
> > > > > >
> > > > > > > Tom,
> > > > > > >
> > > > > > > see below
> > > > > > >
> > > > > > > At 02:49 PM 8/28/2001 -0400, Thomas W. Fredian wrote:
> > > > > > > >Hi Kate,
> > > > > > > >
> > > > > > > > I have succeeded in getting the MDSplus client/server software
> > > > > > > >working using globus_io routines (using version 1.1.3 on
> > > > > > > >portnoyc.lbl.gov, SunOS). It isn't doing any encryption or
> > > > > > > >authentication yet but all the socket I/O is now going through globus.
> > > > > > > >
> > > > > > > > The next step is to add code to have the client authenticate himself
> > > > > > > >to the server. The server needs to check to make sure if the client is
> > > > > > > >allowed to connect and under what account the server should run. I
> > > > > > > >haven't looked much past the globus_io online documentation so I'm not
> > > > > > > >at all sure how this is done. I'll start looking around www.globus.org
> > > > > > > >for some answers but if you have some suggestions they would be
> > > > > > > >appreciated.
> > > > > > >
> > > > > > > Congratulations you are practically done! Since you are using globus io
> > > > > > > security comes for free (to a certain extent that is, I am currently
> > > > > > > investigating if that extent is satisfactory for our purposes;-). What you
> > > > > > > now need to do to add security is essentially set the right attributes on
> > > > > > > your io channel. The best way to go about it is to look at an example for
> > > > > > > globus io included with your globus distribution (let me know if you have
> > > > > > > trouble finding it). and then going by the manual you were using so far
> > > > > > > (globus io) for explanation of functions.
> > > > > > >
> > > > > > > > The server implementation I have running now is in what we call
> > > > > > > >"multi" mode where the server is manually started and it listens on a
> > > > > > > >port and accepts multiple concurrent connections. I'll need to also get
> > > > > > > >it working in single-use mode where it gets run by inetd. It wasn't
> > > > > > > >clear how an application hooks up with an open socket on stdin using the
> > > > > > > >globus_io routines. Any ideas?
> > > > > > >
> > > > > > > I am not sure if I understand your question... It seems like
> > > > > > > globus_io_tcp_connect might perhaps be the answer; if it is not can you
> > > > > > > outline it in more detail?
> > > > > > >
> > > > > > > >Thanks,
> > > > > > > >Tom
> > > > > > >
> > > > > > > __________________________
> > > > > > > Dr. Kate Keahey
> > > > > > > Math & Computer Science Div.
> > > > > > > Argonne National Laboratory
> > > > > > > Argonne, IL 60439, USA
> > > > > > > (630) 252-1673
> > > > >
> > > > > __________________________
> > > > > Dr. Kate Keahey
> > > > > Math & Computer Science Div.
> > > > > Argonne National Laboratory
> > > > > Argonne, IL 60439, USA
> > > > > (630) 252-1673
> > >
> > > __________________________
> > > Dr. Kate Keahey
> > > Math & Computer Science Div.
> > > Argonne National Laboratory
> > > Argonne, IL 60439, USA
> > > (630) 252-1673
> > >
> > > ------------------------------------------------------------------------
> > > Name: io.tar.Z
> > > io.tar.Z Type: Compressed Data (application/x-compress)
> > > Encoding: base64
>
> __________________________
> Dr. Kate Keahey
> Math & Computer Science Div.
> Argonne National Laboratory
> Argonne, IL 60439, USA
> (630) 252-1673
__________________________
Dr. Kate Keahey
Math & Computer Science Div.
Argonne National Laboratory
Argonne, IL 60439, USA
(630) 252-1673
**************************************************************************
**************************************************************************
From: Chris Johnson
Date: Thu, 30 Aug 2001 15:07:57 -0600 (MDT)
To: Kate Keahey
Cc: crj@cs.utah.edu, schissel@fusion.gat.com
Subject: SC01 Fusion demo
Status:
Hi Kate,
Yes, we did discuss this at HPDC. I think we were going to try and do
something simple (since we aren't going to have much in the way of new
research done by then).
We discussed several possibilities, but I don't think we converged on a
specific demo yet.
David - do you remember any specifics?
Cheers,
Chris
Kate Keahey writes:
> Chris,
>
> I seem to recall (and I may be wrong it was rather early in the day ;-) you
> telling me at HPDC that you were planning a visualization demo for the
> Fusion project for SC01. Did I get that right, and if I did what exactly
> were you planning and how is it going to be related to the "remote
> computation" demo? Many thanks!
>
>
> __________________________
> Dr. Kate Keahey
> Math & Computer Science Div.
> Argonne National Laboratory
> Argonne, IL 60439, USA
> (630) 252-1673
**************************************************************************
**************************************************************************
X-Sender: keahey@localhost Date: Thu, 30 Aug 2001 16:06:56 -0500 To:
crj@cs.utah.edu From: Kate Keahey Subject: SC01
Fusion demo Cc: schissel@fusion.gat.com Status:
Chris,
I seem to recall (and I may be wrong it was rather early in the day ;-)
you telling me at HPDC that you were planning a visualization demo for
the Fusion project for SC01. Did I get that right, and if I did what
exactly were you planning and how is it going to be related to the
"remote computation" demo? Many thanks!
__________________________ Dr. Kate Keahey Math & Computer Science Div.
Argonne National Laboratory Argonne, IL 60439, USA (630) 252-1673
**************************************************************************
**************************************************************************
X-Authentication-Warning: apollo.gat.com: majordom set sender to
owner-nfc-workers@fusion.gat.com using -f X-Sender: keahey@localhost
Date: Thu, 30 Aug 2001 16:02:10 -0500 To: "David P. Schissel"
From: Kate Keahey
Subject: Notes on SC01 demo Cc: nfc-workers@fusion.gat.com Sender:
owner-nfc-workers@fusion.gat.com Status:
Here are some belated (very much so) comments on the SC01 demo writeup
that David sent recently.
Comments on demo itself:
EFIT on Show Floor: A Linux workstation would probably be best for ANL.
What does IDL require? Also, did anybody actually volunteer to provide
the GUI?
EFIT, parallel or serial?: let's start with serial EFIT, if we get the
demo to run and have time we can try for parallel
EFIT, how much data?: what are the possible sizes in ballpark? If we
want to have a local "backup" demo version (which I think we should in
case network connectivity goes down) this will translate into storage
resources. Other than that I would say let's try to read and write
whatever amounts of data are necessary to make the story compelling.
MDS events: what we can do is write a module that will wait for an MDS
event and translate it into a Globus callback on the originating
process.
IDEA: It might be good to have a couple of slides explaining in the demo
in detail to show before showing the demo.
"show floor issues"
The idea, I think, was to run the demo in both ANL and LBNL booths.
Posters are a good idea for advertising, but there is a caveats at least
as far as the LANL booth goes. For example, posters must hang wherever
there is space to hang them, and so arranging things so that our poster
hangs over the workstation showing our demo could be hard. Furthermore,
there are some limits on the number of posters as well as the artistic
expression (to give the booth that "unified" look). And they absolutely
have to be done by some company we don't trust, because otherwise things
get too easy. Having said that, I'll do content for our scidac
collaboratories so Fusion will be right up there. Given all this a
better thing for a compelling story will be a well-designed flyer that
people could pick up, keep, and read before or after the demo, and so
far as I know there are no regulations for that.
Getting more data our of EFIT to tell a better story is a great idea.
Having a video would no doubt help tell a compelling story but might be
a lot of effort to make and my concern is that unlike with putting the
actual demo together this effort would be spent entirely on advertising.
Status:
Tom has been making a lot of progress with Globizing MDSplus. I will try
to break down the demo into a list of small components with detailed
description and interfaces and then it would be great to get together
for a telecon and figure out when we need to hit deadlines for specific
components.
__________________________ Dr. Kate Keahey Math & Computer Science Div.
Argonne National Laboratory Argonne, IL 60439, USA (630) 252-1673
===============================================================================
This message was sent to the SciDAC National Fusion Collaboratory (NFC)
workers list nfc-workers.
To unsubscribe from this list, please send a message to
majordomo@fusion.gat.com with the following text in the *body* of the
message: unsubscribe nfc-workers
David P. Schissel:
**************************************************************************
**************************************************************************
X-Sender: keahey@localhost Date: Thu, 30 Aug 2001 16:02:10 -0500 To:
"David P. Schissel" From: Kate Keahey
Subject: Notes on SC01 demo Cc:
nfc-workers@fusion.gat.com Status:
Here are some belated (very much so) comments on the SC01 demo writeup
that David sent recently.
Comments on demo itself:
EFIT on Show Floor: A Linux workstation would probably be best for ANL.
What does IDL require? Also, did anybody actually volunteer to provide
the GUI?
EFIT, parallel or serial?: let's start with serial EFIT, if we get the
demo to run and have time we can try for parallel
EFIT, how much data?: what are the possible sizes in ballpark? If we
want to have a local "backup" demo version (which I think we should in
case network connectivity goes down) this will translate into storage
resources. Other than that I would say let's try to read and write
whatever amounts of data are necessary to make the story compelling.
MDS events: what we can do is write a module that will wait for an MDS
event and translate it into a Globus callback on the originating
process.
IDEA: It might be good to have a couple of slides explaining in the demo
in detail to show before showing the demo.
"show floor issues"
The idea, I think, was to run the demo in both ANL and LBNL booths.
Posters are a good idea for advertising, but there is a caveats at least
as far as the LANL booth goes. For example, posters must hang wherever
there is space to hang them, and so arranging things so that our poster
hangs over the workstation showing our demo could be hard. Furthermore,
there are some limits on the number of posters as well as the artistic
expression (to give the booth that "unified" look). And they absolutely
have to be done by some company we don't trust, because otherwise things
get too easy. Having said that, I'll do content for our scidac
collaboratories so Fusion will be right up there. Given all this a
better thing for a compelling story will be a well-designed flyer that
people could pick up, keep, and read before or after the demo, and so
far as I know there are no regulations for that.
Getting more data our of EFIT to tell a better story is a great idea.
Having a video would no doubt help tell a compelling story but might be
a lot of effort to make and my concern is that unlike with putting the
actual demo together this effort would be spent entirely on advertising.
Status:
Tom has been making a lot of progress with Globizing MDSplus. I will try
to break down the demo into a list of small components with detailed
description and interfaces and then it would be great to get together
for a telecon and figure out when we need to hit deadlines for specific
components.
__________________________ Dr. Kate Keahey Math & Computer Science Div.
Argonne National Laboratory Argonne, IL 60439, USA (630) 252-1673
**************************************************************************
**************************************************************************
Date: Thu, 30 Aug 2001 16:48:06 -0400
From: "Thomas W. Fredian"
X-Accept-Language: en
To: Kate Keahey
CC: Mary Thompson , Martin Greenwald ,
"Schissel, Dave" ,
"McCune, Doug" , Ian Foster
Subject: Re: MDSplus with globus_io
Status:
Hi Kate,
Mary told me of a linux system at lbl I could use. I was able to debug my
problem and I now have the MDSplus mdsip service running using globus
authentication and gsi_wrapping of the packets! I was able to communicate
successfully between a linux system and a sunos system. I used tcpdump on the
packets to convince myself that they were indeed getting encrypted. There is a bit
more work to do to have this part ready for the demo but it seems to be working
fairly well now. Right now the server just prints out the identity of the
credential used to connect to it. It does not validate it or map to a user. This
should be easy to do but shouldn't be critical for the demo. I still will need to
know how one makes a globus service which gets run by inetd (each client gets his
own server process when it issues the connect call). The only other question I had
was whether or not you can alter the handle attributes of a globus connection once
it has been established. I am hoping to gsi_wrap only the mdsplus header portion of
the client server communication and not the whole data transmission. This would
entail switching back and forth between gsi_wrap and clear on the same handle. I
haven't tried this to see if there is a problem doing so.
The problem I was having was caused by code being executed in an unexpected
order. In my server code which accepts connections from clients I was registering a
read callback and then setting up some client data structures. The read callback
was getting executed before my accept code completed. This sure looks like threads
even though I linked to the nothreads globus build libraries. Is there are globus
standard way of creating mutex locks so I can make this thread safe?
Thanks,
Tom
Kate Keahey wrote:
> Tom,
>
> This is wonderful news about MDSplus! We have to be careful that you don't
> solve all of the Fusion Grid problems by SC ;-).
>
> Bug: let's try looking at it under a debugger. If you could request an
> Argonne account I will try to get you the exact software configuration you
> need. f we look under the debugger and nothing happens (nothing simple
> anyways) then the thing to do is to go to the problem reporting page as I
> described in an earlier mail. Send me the number as soon as you get it and
> I'll make sure things get fixed.
>
> For an account: go to https://www-accounts.mcs.anl.gov/ and fill out the
> application form naming me as a sponsor. My signature should contain most
> of the info you'd need (but let me know if that is not enough). This is a
> new thing, we'll see how well it works. It might ask you for resources, say
> "DSL machines". Let me know if the process is annoying.
>
> More below.
>
> At 04:19 PM 8/29/2001 -0400, you wrote:
> >Hi Kate,
> >
> > The examples did help. The MDSplus client is now successfully passing
> >credentials to the server. I'm trying to track down a problem I run into
> >when I
> >turn on GSI_WRAP. With clear mode it all works great. If I turn on
> >GSI_WRAP a few
> >mdsplus messages get sent back and forth successfully but then the server
> >segfaults
> >somewhere. I would love to have an account at Argonne also (preferably on
> >a linux
> >system!). I can't seem to get gdb to work correctly on portnoyc (SunOS) so
> >I've had
> >to debug this thing using brute force (printf's everywhere!).
> >
> > I can now get the identity of the client connecting to the server so
> > I could
> >use this to look up a mapping to a local user using the same code I had in the
> >server for validating user@host. If there is already a globus capability
> >to do this
> >(i.e. given a credential find out if the client is permitted to use a
> >service and
> >which local account that service shoud run under) could you point me to an
> >example?
>
> There is a capability like that: every machine that has Globus installed on
> it contains what is called a gridmap file that maps your credential to a
> local id on that machine. But I do not know if it is accessible through the
> external API. Let me look.
>
> >-tom
> >
> >Kate Keahey wrote:
> >
> > > Tom,
> > >
> > > Here is a tarzipped example; this was the best way for me to learn about
> > > the usage and it should also answer your questions. But please keep the
> > > questions coming if not. Both client and server need to call the functions
> > > to be compatible. In order to obtain credentials you call globus-proxy-init
> > > and that takes credentials from your .globus directory and generates
> > > temporary credentials to use in your session with globus. Would you like to
> > > get an account at Argonne also? It might be easier for me to point you at
> > > examples and reproduce problems if any if you do. Let me know.
> > >
> > > At 04:38 PM 8/28/2001 -0400, you wrote:
> > > >Hi Kate,
> > > >
> > > > I found those routines but there isn't any explanation on how one
> > > > uses them.
> > > >Having no experience dealing with secure communications it isn't very
> > > >clear who
> > > >needs to call these (client or server), when and what to use as arguments. For
> > > >example, who calls globus_io_attr_set_secure_authentication_mode? and what
> > > >do I
> > > >use for the credential argument? I'm not sure where to find a globus_io
> > > >example.
> > > >I'm working with this on a sun system at lbl.
> > > >
> > > >-tom
> > > >
> > > >Kate Keahey wrote:
> > > >
> > > > > OK, let me look into your question.
> > > > >
> > > > > For Globus io documentation look at the attribute functions. You can search
> > > > > on "set_secure_authentication_mode", and there they all are more or less.
> > > > > Also channel wrapping. And let me know if you encounter any problems at all.
> > > > >
> > > > > At 04:03 PM 8/28/2001 -0400, you wrote:
> > > > > >I'll look some more at the globus_io documentation. globus_io_tcp_connect is
> > > > > >used by the client to connect to a server. What I need is some mechanism for
> > > > > >the server to begin using globus_io on stdin (which is the socket connection
> > > > > >to the client when the application is run by inetd).
> > > > > >
> > > > > >-tom
> > > > > >
> > > > > >Kate Keahey wrote:
> > > > > >
> > > > > > > Tom,
> > > > > > >
> > > > > > > see below
> > > > > > >
> > > > > > > At 02:49 PM 8/28/2001 -0400, Thomas W. Fredian wrote:
> > > > > > > >Hi Kate,
> > > > > > > >
> > > > > > > > I have succeeded in getting the MDSplus client/server software
> > > > > > > >working using globus_io routines (using version 1.1.3 on
> > > > > > > >portnoyc.lbl.gov, SunOS). It isn't doing any encryption or
> > > > > > > >authentication yet but all the socket I/O is now going through globus.
> > > > > > > >
> > > > > > > > The next step is to add code to have the client authenticate himself
> > > > > > > >to the server. The server needs to check to make sure if the client is
> > > > > > > >allowed to connect and under what account the server should run. I
> > > > > > > >haven't looked much past the globus_io online documentation so I'm not
> > > > > > > >at all sure how this is done. I'll start looking around www.globus.org
> > > > > > > >for some answers but if you have some suggestions they would be
> > > > > > > >appreciated.
> > > > > > >
> > > > > > > Congratulations you are practically done! Since you are using globus io
> > > > > > > security comes for free (to a certain extent that is, I am currently
> > > > > > > investigating if that extent is satisfactory for our purposes;-). What you
> > > > > > > now need to do to add security is essentially set the right attributes on
> > > > > > > your io channel. The best way to go about it is to look at an example for
> > > > > > > globus io included with your globus distribution (let me know if you have
> > > > > > > trouble finding it). and then going by the manual you were using so far
> > > > > > > (globus io) for explanation of functions.
> > > > > > >
> > > > > > > > The server implementation I have running now is in what we call
> > > > > > > >"multi" mode where the server is manually started and it listens on a
> > > > > > > >port and accepts multiple concurrent connections. I'll need to also get
> > > > > > > >it working in single-use mode where it gets run by inetd. It wasn't
> > > > > > > >clear how an application hooks up with an open socket on stdin using the
> > > > > > > >globus_io routines. Any ideas?
> > > > > > >
> > > > > > > I am not sure if I understand your question... It seems like
> > > > > > > globus_io_tcp_connect might perhaps be the answer; if it is not can you
> > > > > > > outline it in more detail?
> > > > > > >
> > > > > > > >Thanks,
> > > > > > > >Tom
> > > > > > >
> > > > > > > __________________________
> > > > > > > Dr. Kate Keahey
> > > > > > > Math & Computer Science Div.
> > > > > > > Argonne National Laboratory
> > > > > > > Argonne, IL 60439, USA
> > > > > > > (630) 252-1673
> > > > >
> > > > > __________________________
> > > > > Dr. Kate Keahey
> > > > > Math & Computer Science Div.
> > > > > Argonne National Laboratory
> > > > > Argonne, IL 60439, USA
> > > > > (630) 252-1673
> > >
> > > __________________________
> > > Dr. Kate Keahey
> > > Math & Computer Science Div.
> > > Argonne National Laboratory
> > > Argonne, IL 60439, USA
> > > (630) 252-1673
> > >
> > > ------------------------------------------------------------------------
> > > Name: io.tar.Z
> > > io.tar.Z Type: Compressed Data (application/x-compress)
> > > Encoding: base64
>
> __________________________
> Dr. Kate Keahey
> Math & Computer Science Div.
> Argonne National Laboratory
> Argonne, IL 60439, USA
> (630) 252-1673
**************************************************************************
**************************************************************************
