Josh,
Great. Thanks for the clear explanation. I'll touch base with Kelly here at GSU and see what status is of changes she is working on implementing the VCL 2.2.1


Appreciate the assist.


Art

On Jun 6, 2011, at 11:17 AM, Josh Thompson wrote:

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

First, for anyone who hasn't heard something about the "VCL office hours", there is a conference call scheduled every other Tuesday afternoon (4-5pm EST) organized and hosted by Andy Rindos from IBM. This call is not officially connected with the Apache Software Foundation. Andy has been kind enough to
host the call for various people interested in using VCL and sharing
information about their experiences via a voice conference call.

Art,

I'm pretty sure I know what is happening with your system. Hopefully, I can
communicate it clearly.

VCL keeps any unused machines in a "hot standby" state. That means it deploys the unused machines with some image up until the point of needing to add a user account. It uses a predictive algorithm to select what image to load on a machine at the end of each reservation. The idea is to get through most of the deploy procedure before someone requests the image so they only have to wait for a small portion of it (adding the user account and setting up access rules) when they actually make a reservation. Any time you see an estimated load time of less than one minute, you are hitting a hot standby machine. If there are no machines that have been pre-deployed with the image, you get a longer estimated load time, in which case the estimated load time is an
average of the last several times the image was deployed.

So, at the end of every reservation, an image gets deployed to the machine
that was used.

The last time we talked about the the GSU setup, you were using the ESX deployment module and had a storage array that was somewhat slow. I haven't heard anything about that being changed. The ESX deployment module makes a full copy of the image disk files for every deploy. So, when you are ending your reservation, you are triggering a new deploy on the machine you used, which includes a full copy of the disk files on your slow storage array. If you go ahead and make a new reservation before that deploy finishes, you are now doing a second full copy of the disk files on your slow storage array, which is competing with the first one, making both copies take longer. If you wait the 5 or so minutes you referred to, then you don't slow down the deploy
time by introducing a second deploy to compete with it.

This problem should go away when you switch to using the VMware deployment module that doesn't do full copies at every deploy (it uses VMware's non- persistent mode when it can) and also when you get a faster storage array.

I hope that clears things up. Let me know if you'd like me to clarify any of
it.

Josh

On Thursday June 02, 2011, Art Vandenberg wrote:
From the VLC office hours on Tuesday past, I promised to follow up on
problem I experienced with New Reservation when I'd ENDed a prior
reservation.

DISCLAIMER - this was my experience here at GSU, your experience may
be different.

SCENARIO 1:
a) My VCL Matlab session comes to 10 minute warning.
b) I stop Matlab, logged off, and then END Current reservation (which
still has not reached end time). NOTE: assume I can not extend past
our limit.
c) I immediately request NEW (NOW) reservation; message indicates
"Estimated Load Time < 7 minutes."
d) I back off and wait 2 minutes; retry NEW reservation, still
"Estimated Load Time  < 7 minutes..."
e) I back off for 5 minutes and when I select NOW reservation, the
create time shows "Estimated Load Time < 1 minute..." and reservation
made in < 1!

SCENARIO 2:
I subsequently tested same scenario, but this time with a reservation
that was ENDed but still had nearly 55 minutes remaining... and same
result: if I wait 5+ minutes, the "Estimated Load Time < 1" (and the
message appears to switch to "< 7 minutes" right at the 5 minute mark
after ENDing a CURRENT reservation.)

SCENARIO 3:
And I then tested what happens if I just go ahead and immediately
request the NOW reservation (after ENDing a CURRENT reservation) and
ignore the "Estimated Load Time  < 7 minutes...".  The Pending window
was open so I could see progress as shown following:

PENDING WINDOW SNAPSHOT AT 8:08
State Est/Act
Time Total
Time
confirming image exists(22) 0:04/0:07 0:07
starting load process(40) 0:06/8:01 8:08
transferring image files to host server(27) 4:15
creating configuration file(28) 0:02
starting virtual machine(48) 0:03
machine booting(46) 1:08
running post configuration(9) 0:44
resource ready(43) 0:00
adding user account(41) 0:18
ready for Connect(18) 0:00

PENDING WINDOW SNAPSHOT AT 8:23
State Est/Act
Time Total
Time
confirming image exists(22) 0:04/0:07 0:07
running post configuration(9) 0:50/8:16 8:23
resource ready(43) 0:00/0:26 8:49
adding user account(41) 0:18
ready for Connect(18) 0:00

PENDING WINDOW SNAPSHOT AT 9:09
State Est/Act
Time Total
Time
confirming image exists(22) 0:04/0:07 0:07
running post configuration(9) 0:50/8:16 8:23
resource ready(43) 0:00/0:45 9:08
adding user account(41) 0:18/0:01 9:09
ready for Connect(18) 0:00


PENDING WINDOW SNAPSHOT AT 9:12
State Est/Act
Time Total
Time
confirming image exists(22) 0:04/0:07 0:07
running post configuration(9) 0:55/8:16 8:23
resource ready(43) 0:00/0:45 9:08
ready for Connect(18) 0:00/0:04 9:12


As you see, the "starting load process" took 8:08 minutes to complete
and image was ready only at 9:12.


Basically, if I END a reservation before its END time, I should wait
5 minutes before requesting NEW reservation.
---> I don't need MATLAB to see that (wait 5 minutes + less than 1
minute for NEW image) IS LESS THAN (don't wait + 9:12 minutes)!

Whether that argues for a forced wait time in the code (if user has
ENDed reservation, enforce 5 minute wait before allowing NEW request),
I leave to others.

NOTE: Kelly R. advise if you've maybe already adjusted code here at GSU!


Later



Art

Art Vandenberg
Account Manager/Research Function
Customer Relations, IS&T
Information Systems & Technology
Georgia State University
avandenb...@gsu.edu
+1 404 413 4743
MS Information & Computer Science, Georgia Tech
MVA Painting & Drawing, Georgia State
Web page: http://www.gsu.edu/ist/acs/25735.html
- --
- -------------------------------
Josh Thompson
VCL Developer
North Carolina State University

my GPG/PGP key can be found at pgp.mit.edu
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (GNU/Linux)

iEYEARECAAYFAk3s74wACgkQV/LQcNdtPQNLdgCfXLUBKkMHMTMmO1LyUBbmOQiQ
HU4An0+f6x3phzqqU4SsJXN4+I4zl+Ue
=G5AA
-----END PGP SIGNATURE-----

Art Vandenberg
Account Manager/Research Function
Customer Relations, IS&T
Information Systems & Technology
Georgia State University
avandenb...@gsu.edu
+1 404 413 4743
MS Information & Computer Science, Georgia Tech
MVA Painting & Drawing, Georgia State
Web page: http://www.gsu.edu/ist/acs/25735.html



Reply via email to