INTEL CORPORATION

Patent Trials and Appeals Board

Oct 13, 2020

2020001059 (P.T.A.B. Oct. 13, 2020)

UNITED STATES PATENT AND TRADEMARK OFFICE
UNITED STATES DEPARTMENT OF COMMERCE
United States Patent and Trademark Office
Address: COMMISSIONER FOR PATENTS
P.O. Box 1450
Alexandria, Virginia 22313-1450
www.uspto.gov
APPLICATION NO. FILING DATE FIRST NAMED INVENTOR ATTORNEY DOCKET NO. CONFIRMATION NO.
15/529,426 05/24/2017 Yaozu Dong P69781US 1132
96162 7590 10/13/2020
Law Office of R. Alan Burnett, PS
c/o CPA Global
900 Second Avenue South, Suite 600
Minneapolis, MN 55402
EXAMINER
LIU, GORDON G
ART UNIT PAPER NUMBER
2612
NOTIFICATION DATE DELIVERY MODE
10/13/2020 ELECTRONIC
Please find below and/or attached an Office communication concerning this application or proceeding.
The time period for reply, if any, is set in the attached communication.
Notice of the Office communication was sent electronically on above-indicated "Notification Date" to the
following e-mail address(es):
alan@patentlylegal.com
docketing@cpaglobal.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
____________________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________________

Ex parte YAOZU DONG and KUN TIAN
____________________

Appeal 2020-001059
Application 15/529,426
Technology Center 2600
____________________

Before JOSEPH L. DIXON, ROBERT E. NAPPI, and,
JASON J. CHUNG, Administrative Patent Judges.

NAPPI, Administrative Patent Judge.

DECISION ON APPEAL
Appellant1 appeals under 35 U.S.C. § 134(a) from the Examiner’s
final rejection of claims 26 through 50. We have jurisdiction under
35 U.S.C. § 6(b).
We reverse.

1 We use the word Appellant to refer to “applicant” as defined in 37 C.F.R.
§ 1.42(a). According to Appellant, Intel Corporation, is the real party-in-
interest. Appeal Br. 3.
Appeal 2020-001059
Application 15/529,426

2
CLAIMED SUBJECT MATTER
The claims relate generally to high availability Graphical Processor
Unit (GPU) computing with virtual machine checkpointing. Spec. Abstract.
Claim 26 is reproduced below.
26. A method comprising
implementing at least one virtual machine (VM) on a
compute platform including a central processing unit (CPU), a
graphics processing unit (GPU), and graphics memory, each of
the at least one virtual machine hosted by a hypervisor executed
via the CPU;
for each of at least one VM,
trapping GPU commands submitted from the VM;
emulating, using a virtual GPU associated with the
VM, changes to state information for the GPU that are
predicted to result when the trapped GPU commands are
executed by the GPU;
predicting graphics memory pages that might be
dirtied via execution of trapped GPU commands by the
GPU; and
periodically performing a VM checkpointing
operation for the VM, wherein a snapshot of changes to
the state information for the GPU and a copy of the
graphics memory pages that are predicted to-be-dirtied
are stored as a VM checkpoint.

EXAMINER’S REJECTIONS2
The Examiner rejects claims 26 through 42, and 45 through 50 under

2 Throughout this Decision we refer to the Appeal Brief filed June 26, 2019
(“Appeal Br.”); Reply Brief filed November 20, 2019 (“Reply Br.”); Final
Office Action mailed January 30, 2019 (“Final Act.”); and the Examiner’s
Answer mailed September 26, 2019 (“Ans.”).
Appeal 2020-001059
Application 15/529,426

3
35 U.S.C. § 103 as unpatentable over the combined teachings of Stone (US
2011/0134111 A1), Hiltgen (US 2008/0155169 A1) and Yin (US
2013/0185716 A1). Final Act. 8–33.
The Examiner rejects claims 43 and 44 under 35 U.S.C. § 103 as
unpatentable over the combined teachings of Stone, Hiltgen Yin and Earl
(US 9,201,887 B1). Final Act. 33–35.
ANALYSIS
We have reviewed Appellant’s arguments in the Appeal Brief, the
Examiner’s rejections, and the Examiner’s response to Appellant’s
arguments. Appellant’s arguments have persuaded us of error in the
Examiner’s rejection of all of the claims under 35 U.S.C. § 103.
Independent claim 26
Appellant presents several arguments directed to the Examiner’s
obviousness rejection of claim 26. Appeal Br. 14– 37. The dispositive issue
presented by these arguments is did the Examiner err in finding the
combination of Stone, Hiltgen and Yin teach the claim 26 limitations of
“predicting graphics memory pages that might be dirtied via execution of
trapped GPU commands” and “a snapshot of changes to the state
information of the GPU and a copy of the graphics memory pages that are
predicted to-be-dirtied are stored as a VM checkpoint.” Appellant
summarizes Stone as directed to a scheme for remote rendering of three
dimensional images using virtual machines in which a hypervisor allocates
to a virtual machine, exclusive direct asses to a GPU. Id. at 18. A second
virtual machine intercepts three dimension draw commands generated by an
application and forwards the intercepted commands to the first virtual
machine to which the GPU is allocated. Id. 18–19. Appellant summarizes
Appeal 2020-001059
Application 15/529,426

4
Hiltgen as directed to implementing a virtual machine using storage system
functionality, where the state data for the virtual machine is represented on
the storage system. Id. at 19. Appellant asserts that there is no mention of a
GPU in Hiltgen and that a virtual machine is not analogous to emulating a
GPU. Id. at 19 and 23. Further, Appellant argues that the prediction aspect
of claim 26 is not taught by Hiltgen. Id. at 24. Appellant argues:
migration of a VM involves migrating its current state,
which would include that current state of memory pages
allocated by the VMM or Hypervisor for use by the VM.
The dirty pages reflect pages that have actually been
dirtied (i.e., changed), as opposed to pages that might be
dirtied via execution of trapped GPU commands by the
GPU. The reason for tracking dirtied pages under
Hiltgen is so that pages that have been allocated to a VM
but have yet to be used (and thus not dirtied) do not need
to be paged out. There would be no point in storing data
for such unused pages. Again, these pages aren’t
predicted to be dirtied (for any reason, much less via
execution of trapped GPU commands by a (non-existent)
GPU) under Hiltgen, but are, in fact, dirtied pages.
An Examiner interview was conducted between
the undersigned representative and the Examiner on
March 12, 2019 to discuss the rejection of the claims in
the Final Office Action. In particular, the rejection of
claim elements including the limitation, “predicting
graphics memory pages that might be dirtied via
execution of trapped GPU commands by the GPU” were
discussed. The Examiner acknowledged that Hiltgen’s
teaching of tracking memory pages and checking whether
pages are dirtied or not does not teach this claim
limitation, as predicting pages that might be dirtied
would be done before the pages would be dirtied (in
advance of the pages being dirtied), while tracking pages
to check whether the page(s) is/are dirtied could only be
done after the page(s) is/are dirtied.

Appeal 2020-001059
Application 15/529,426

5
Appeal Br. 26–27 (referring to Hiltgen ¶ 110).
The Examiner responds to Appellant’s arguments, concerning the
dispositive issue, stating:
Hiltgen teaches that the state data is paged out and there
is an indication if the page includes new data. These data
are sent to the storage unit, and the destination
virtualization layer is monitoring the storage layer; in
addition, these data may be sent to the destination
virtualization layer directly. If these data reach a limit,
the destination virtualization layer may request new
pages. Whether the destination virtualization layer
requests new page or not depend on the new data (See
Hiltgen: Fig. 16, and [0110], “For example, dirty pages
may be tracked and re-synchronized iteratively. The
source virtualization system also indicates if a page
includes new data in the network storage unit. Instead of,
or in addition to, writing an indication into the network
storage unit, the source may inform the destination
directly, (e.g., over an IP network). If the virtualization
layer of the destination virtualization system detects the
new data indication, then it requests the one or more new
pages. Requested pages are then loaded into the
destination virtualization system by the virtualization
layer. These operations repeat until a given limit is
achieved”), and this feature may be corresponding to the
“prediction” feature claimed in the claim 26.

Answer 4–5.
We disagree with the Examiner’s finding that Hiltgen teaches the
claimed prediction. Claim 26 recites “predicting graphics memory pages that
might be dirtied via execution of trapped GPU commands by the GPU.”
Thus, the claimed prediction of pages that may (i.e. looking to the future) be
dirtied, not pages that have been (looking to the past) dirtied (changed
memory), we do not consider repeating the detecting a new data indication
Appeal 2020-001059
Application 15/529,426

6
to be predicting dirtied memory, rather it appears to be an indication that
memory is dirtied. Thus, we do not consider the Examiner to have shown
that the combination of Stone, Hiltgen and Yin teaches all of the limitations
of independent claim 26, and we do not sustain the rejection of independent
claim 26 and claims 27 through 34 which depend upon claim 26.

Independent claims 35 and 46
Appellant presents several arguments directed to the Examiner’s
obviousness rejection of claim 35 and 46. Appeal Br. 47-52 and 55. The
dispositive issue presented by these arguments is did the Examiner err in
finding the combination of Stone, Hiltgen Yin teach the claim 35 and 46
limitations directed to “trap[ing] access to predetermined GPU resources
made by the graphics driver.” Id. 46–53, 55. Appellant argues that the only
thing trapped in Stone are the three dimensional drawing commands and not
the resources access to GPU resources. Id. at 49–50 (citing Stone ¶ 60).
The Examiner does not persuasively respond to Appellant’s specific
arguments regarding claims 35 and 46 other than that the rejection is proper
for the reasons discussed in response to the rejection of claim 26. In the
Final Action, the Examiner cites to Stone paragraph 90 as teaching trapping
commands from the graphics driver and to Stone paragraph 60 as teaching
trapping GPU resources. Final Act. 18.
We are persuaded of error in the Examiner’s rejection by Appellant’s
arguments. Initially, we note that independent claims 26 is of different
scope than claims 35 and 46, in that claim 26 does not recite trapping
resources of the GPU as recited in claims 35 and 46. Thus, the Examiner’s
response to the Appellant’s arguments concerning claim 26 do not directly
Appeal 2020-001059
Application 15/529,426

7
apply to this limitation. Further, we have reviewed the paragraphs of Stone
cited by the Examiner in the Final Action, and we do not find that they
discuss trapping resources of the GPU nor has the Examiner explained how
the cited paragraphs teach or suggest trapping resources. Thus, we do not
consider the Examiner to have persuasively shown that the combination of
Stone, Hiltgen and Yin teaches all of the limitations of independent claims
35 and 46. Accordingly, we do not sustain the rejection of independent
claim 35 and 46, or the rejection of 36 through 42, 45, and 47 through 50
which depend upon claims 35 and 46.
The Examiner’s rejection of claims 43 and 44, similarly rely upon the
teachings of Stone, Hiltgen, and Yin to teach the limitations of independent
claim 34 upon which they ultimately depend. Final Act. 33. Thus, we
similarly do not sustain the Examiner’s rejection of these claims for the
same reasons as discussed with respect to claim 35.
CONCLUSION
In summary:
Claim
Rejected
35 U.S.C. § Reference(s)/
Basis
Affirmed Reversed
26–42, 45–
50
103 Stone,
Hiltgen, Yin
26–42, 45–50
43, 44 103 Stone,
Hiltgen, Yin,
Earl
43, 44
Overall
Outcome
26–50
REVERSED