PhD defense
-
Upload
chuckcartledge -
Category
Education
-
view
468 -
download
2
description
Transcript of PhD defense
A Framework for Web Object Self-Preservation
A Ph.D. Defense
Chuck Cartledge30 May 2014
2
A warning from Jeff Rothenberg
“Digital Information Lasts Forever—Or Five Years, Whichever Comes First.”
Jeff Rothenberg, Ensuring the Longevity of Digital Information, Scientific American 272 (1995), 42 - 47.
3
A warning from William Arms
“Tomorrow we could see the National Library of Medicine abolished by Congress, Elsevier dismantled by a corporate raider, the Royal Society declared bankrupt, or the University of Michigan Press destroyed by a meteor. All are highly unlikely, but over a long period of time unlikely events will happen."
William Y. Arms, Preservation of Scientific Serials: Three Current Examples, Journal of Electronic Publishing 5 (1999), no. 2.
4
Overview
• Warnings• Preservation context• Research questions• Background and related work• Unsupervised Small-World– Emergent behavior– Graph theory– Preservation
• Demonstration• Questions and answers
5
Preservation in an analog age
• Benign neglect– Don’t touch– Keep away from sunshine– Keep away from moisture– Keep away from insects
• Last for hundreds of years
Josie McClure picture taken Feb 30, 1907 at Poteau, I.T. Fifteen years of age When this was taken weighed 140 lbs.
6
Preservation in a digital age
• Constant use– Use often– Exposure to lots of things– Make lots of copies– Monitor the integrity
• Last for ??? unknown years
• This is a Brave New World
Google image search, 31 March 2014, about 91,700,000 results (0.84 seconds)
7
Everything has a lifespan
• Exponential growth of digital artifacts
• Representing increasing portion of personal and cultural heritage
• Short human lifetime to manage data
• Potentially, short institutional life time
• Need to preserve artifacts beyond human lifespan and institutions that create and house artifacts
Dissertation, section 1.3
8
Research questions
• Can web objects (WOs) be constructed to outlive the people and institutions that created them?
• Can we leverage aspects of naturally occurring networks and group behavior for preservation?
A WO is a digital object that lives on the Web. A WO is a fundamental element in this dissertation.
9
Unsupervised Small-World (USW) is at the nexus of multiple disciplines
Mathematical structures used to
model pairwise relations between
objects
Ensuring that digital information
of continuing value remains accessible and
usable
Movement of the inanimate
Shifting gears
10
Focusing on emergent behavior
11
Emergent behavior: model• Craig Reynolds – basis of herd
and flock behavior in computer animations– 3 rules
• Collision avoidance• Velocity matching• Flock centering
– No central control, everything based on local knowledge only
• Simple rules– Complex behavior– Emergent behavior Craig W. Reynolds, Computer Animation with Scripts and
Actors, ACM SIGGRAPH, vol. 16, ACM, 1982, pp. 289 - 296. Images http://www.red3d.com/cwr/boids/ Flock centering
Velocity matching
Collision avoidance
12
Emergent behavior: communication
• Need to know what my neighbors are doing
• Need to tell neighbors what I am doing
• A school of fish do not have a Dagon that controls them
Dissertation, section 5.3
Shifting gears
13
Focusing on preservation
14
Preservation: primitives
William Y. Arms, Digital Libraries, The MIT Press, December 1999
png
png png pngReplication Emulation
png
tiff eps bmpMigration
15
Preservation: OAIS model• Provides standard
model and terminology for archival systems
• Terms of interest– Submission
Information Package
– Ingest– Data
Management– Archival Storage– Access– Dissemination
Information Package Council of the Consultative Committee for Space Data Systems (CCSDS), Reference Model for an Open
Archival Information System (OAIS), Tech. report, Consultive Committee for Space Data Systems 650.0-M-2, Magenta Book, 2012.
Carl Lagoze, Herbert Van de Sompel, Pete Johnston, Michael Nelson, Robert Sanderson, and Simeon Warner, ORE User Guide - Resource Map Implementation in Atom, Tech. report, Open Archives Initiative, 2004.
Shifting gears
16
Focusing on graph theory
17
Graph theory: definitions
• Graph: G = (V,E)• Graph can be connected or
disconnected• Some graph metrics work only
with connected graphs and not with disconnected graphs– Clustering coefficient (C(G))– Average path length (L(G))– Degree distribution
Reka Albert and Albert-Laszlo Barabasi, Statistical Mechanics of Complex Networks, Reviews of Modern Physics 74 (2002), no. 1, 47.
18
Graph theory: Watts and Strogatz small-world
Duncan J. Watts and Steven H. Strogatz, Collective dynamics of `small world' networks, Nature 393 (1998), 440 - 442.
Stanley Milgram, The Small-World Problem, Psychology Today 2 (1967), no. 1, 60 - 67.
19
Small-world graphs are common
Actual Random graph
Nodes Edges C(G) L(G) C(G) L(G)WECC 4941 6594 0.0801 18.99 0.00054 8.7
C. elegans
248 511 0.21 2.87 0.05 2.62
Email 148 ~500,000 0.44 2.25 0.11 2.0
Ake J Holmgren, Using Graph Models to Analyze the Vulnerability of Electric Power Networks, Risk Analysis 26 (2006), no. 4, 955 - 969.
Lav R Varshney, Beth L Chen, Eric Paniagua, David H Hall, and Dmitri B Chklovskii, Structural Properties of the Caenorhabditis elegans Neuronal Network, PLoS computational biology 7 (2011), no. 2, e1001066.
Shinako Matsuyama and Takao Terano, Analyzing the ENRON Communication Network Using Agent-Based Simulation, Journal of Networks 3 (2008), no. 7.
West Elect. Coord. Council Enron
20
Small-world: high C(G) and low L(G)
• The ubiquitous presence of small-world graphs points to something inherently “correct” and desirable about them.
Symbol Meaning
k Degree
n Order of the graph
Dissertation, section 5.2.3
21
USW is at the nexus of multiple disciplines
Creation of small-world
graphs that are robust and
resilient
Meet fundamental requirements of replication, migration, and
data management
WO’s use of emergent
behavior to create, monitor,
and optimize the USW system
22
Euclidean geometry• To draw a straight line from any point to
any point.• To produce [extend] a finite straight line
continuously in a straight line.• To describe a circle with any center and
distance [radius].• That all right angles are equal to one
another.• That, if a straight line falling on two straight
lines make the interior angles on the same side less than two right angles, the two straight lines, if produced indefinitely, meet on that side on which are the angles less than the two right angles.
The sum of angles A, B, and C is equal to 180 degrees
Euclid of Alexandria, The Elements, Alexandria, 300 BCE.
23
Non-Euclidean geometries• Non-Euclidean geometry
– To draw a straight line from any point to any point.
– To produce [extend] a finite straight line continuously in a straight line.
– To describe a circle with any center and distance [radius].
– That all right angles are equal to one another.– That, if a straight line falling on two straight
lines make the interior angles on the same side less than two right angles, the two straight lines, if produced indefinitely, meet on that side on which are the angles less than the two right angles.
• Spherical geometry– Two lines at right angles to the same line can
meet– Triangles can have 180 to 540 degrees– Circles are straight lines
24
Digital library world• Digital libraries
– The technical framework exists within a legal and social framework
– Understanding of digital library concepts is hampered by terminology
– The underlying architecture should be separate from the content stored in the library
– Names and identifiers are the basic building block for the digital library
– Digital library objects are more than collections of bits
– The digital library object that is used is different from the stored object
– Repositories must look after the information they hold
– Users want intellectual works, not digital objects
• Basic digital library tenets
Robert Kahn and Robert Wilensky, A Framework for Distributed Digital Object Services, International Journal on Digital Libraries 6 (2006), no. 2, 115 - 123.
William Y. Arms, Key Concepts in the Architecture of the Digital Library, D-Lib Magazine 1 (1995), no. 1.
25
Digital library worlds of possibilities
• Digital libraries– The technical framework exists within a legal and
social framework– Understanding of digital library concepts is hampered
by terminology– The underlying architecture should be separate from
the content stored in the library– Names and identifiers are the basic building block for
the digital library– Digital library objects are more than collections of
bits– The digital library object that is used is different from
the stored object– Repositories must look after the information they
hold– Users want intellectual works, not digital objects
What if there were no repositories?
26
“No Repositories” → USW
• No global knowledge– No omnipotent
enforcer– No omnipresent
monitor• Opportunistic
preservation• Self-describing
Web Objects
USW contributions
27
28
USW WO “friendship” links • WOs have
“friendship” links to other WOs
• Different than HTML navigational links
Dissertation, Chapter 6
29
USW WO “families”
A family is a set of copies of the same WO
Dissertation, Chapter 6
30
USW hosts Family members live on different hosts Host #1
Host #2
Host #3 Dissertation, Chapter 6
Shifting gears
31
Focus on emergent behavior
32
USW interpretation of flocking
Flock centering
Velocity matching
Collision avoidance
Craig Reynolds’ “boids” USW interpretation
Each WO has a unique URI
Matching number of copies/family members
Move with friends to new hosts
Dissertation, Chapter 2
33
Building a USW graph
• Graph exploration ( )b
• Choosing connections
• Detecting loss
Dissertation, Chapter 5
34
WOs wandering in the USW graph
• Wandering WO is “introduced” to an existing WO
• If a connection is not made, then an attempt is made to another existing WO
• Process is repeated until a connection is made
• No global knowledge– No omnipotent
enforcer– No omnipresent
monitor• No repositories
Dissertation, Chapter 5
35
USW friend selection process
• Selection from possible sets– WOset: WOs connected to candidate WO– visitedSet: WOs that the wandering WO has explored– toBeVisitedSet: WOs that the wandering WO has
discovered• Selection approaches
– Random from visitedSetUtoBeVisitedSet – FIFO from visitedSetUtoBeVisitedSet – LIFO from visitedSetUtoBeVisitedSet – Preferentially attach to WOset then random for remaining
Dissertation section 6.7.5
36
Comparing USW to random graphs
Dissertation, section 6.4
Shifting gears
37
Focus on graph theory
38
Robustness of USW graphs
• Definition: able to continue when damaged
• Attack vs. failure– Intentional vs. random
• Component selection– Vertex– Edge
• Selection attribute– Degree– Betweeness
• Attribute value– High– Low
• Attack profile notation: A{D|E|V}{H|L}
Sample graph
Charles L. Cartledge and Michael L. Nelson, Connectivity Damage to a Graph by the Removal of an Edge or Vertex, Tech. report, arXiv 1103.3075, ODU CS Dept.
39
Different attack profiles selections
AEH AEL AVH
ADLAVL ADH
Charles L. Cartledge and Michael L. Nelson, Connectivity Damage to a Graph by the Removal of an Edge or Vertex, Tech. report, arXiv 1103.3075, ODU CS Dept.
40
Four attacks using AEL profile
Deletion #1. Deletion #2.
Deletion #3. Deletion #4.
41
Our damage vs. Albert, Jeong, and Barabasi’s damage
s = Reka Albert, Hawoong Jeong, and Albert-Laszlo Barabasi, Error and Attack Tolerance of Complex Networks, Nature 406 (2000), no. 6794, 378 - 382.
50 … 5 20 … 20
16 … 1 10 … 10
42
Measuring damage
• Desired characteristics– Different framgentation cases result
in different values– Useful without additional graph
state information
Charles L. Cartledge and Michael L. Nelson, Connectivity Damage to a Graph by the Removal of an Edge or Vertex, Tech. report, arXiv 1103.3075, ODU CS Dept.
Dissertation Appendix D
43
Local and global AE* damage
44
Global A{DV}H damage (100 nodes)
• 100 node graph• Execution time: ~36
hours• Attacker has total
knowledge of the graph
• Attacker has unrestricted resources to damage the graph
• Results:– Small-world the
most connected is not the most valuable
– Random and scale-free degreeness does not make a difference
45
Attack profile efficacy on sample graph
Attack profile Attacks efficacy
AEdge High The core of the graph 1.43
AEdge Low The periphery of the graph 1.00
AVertex High The core of the graph 1.42
AVertex Low The periphery of the graph 1.00
ADegree High The core of the graph 1.40
ADegree Low The periphery of the graph 1.00
• If the attacker's goal is to disconnect the graph by repeated use of the same attack profile, then the most effective profiles in order are: AEH , AVH , and ADH.
• HTTP/HTML does not support AE* attack profiles Dissertation, section 5.6.6
46
Detecting loss of family members
• Each “active maintainer” WO checks its family’s status– Check family member
accessibility– Check friend accessibility
• If family member is lost, use friends to select candidate host
• If too few candidate hosts, use friends to explore and discover new hosts
Dissertation, section 6.8
Shifting gears
47
Focus on preservation
48
When to make family members?
• What is a copy?• Who makes the copies?• How many to make? Answer:
defined by originating domain– 0 to start– Soft lower limit (csoft)
– Hard upper limit (chard) • Where to make them?
– Distributed across known hosts– Too many or too few hosts
• When to make them?
Norman Paskin, On Making and Identifying a Copy, D-Lib Magazine 9 (2003), no. 1. Henry M. Gladney and John L. Bennett, What Do We Mean by Authentic? What's the
Real McCoy?, D-Lib Magazine 9 (2003), no. 7/8.
49
USW preservation definitions• Hierarchy of family WOs
– Progenitor – initial WO– Copies – more recent WO copies– Each WO is timestamped with creation time
• WO roles– Active maintainer – eldest WO charged with
making copies and related housekeeping– Passive maintainer – all other WOs
• Order of precedence– If progenitor is accessible then it is the active
maintainer– If declared active maintainer is accessible then it is
the active maintainer– Otherwise, WO declares itself active maintainer
• If family is disconnected then multiple active maintainers are possible until reconnection then the eldest WO declares itself active maintainer
Progenitor
Copies
Dissertation, Appendix A
50
Active and passive maintenance activities
Active
Passive
Active Active
PassivePassive
• Active maintainer (the WO with earliest timestamp) – currently charged with making copies and related housekeeping
• Passive maintainer – all other WOs
XCopy declares
Itself a
ctive maintainer
ProgenitorIs lost
Progenitorreturns
Progenitordeclares
act. as copy.
Time
51
Progenitor is lostActive
Passive
Active Active
PassivePassive
• Active maintainer – currently charged with making copies and related housekeeping
• Passive maintainer – all other WOs
XCopy declares
Itself a
ctive maintainer
ProgenitorIs lost
Progenitorreturns
Progenitordeclares
act. as copy.
Time
52
A new active maintainerActive
Passive
Active Active
PassivePassive
• Active maintainer – currently charged with making copies and related housekeeping
• Passive maintainer – all other WOs
XCopy declares
Itself a
ctive maintainer
ProgenitorIs lost
Progenitorreturns
Progenitordeclares
act. as copy.
Time
53
Progenitor returns and assumes active maintainer role
Active
Passive
Active Active
PassivePassive
• Active maintainer – currently charged with making copies and related housekeeping
• Passive maintainer – all other WOs
XCopy declares
Itself a
ctive maintainer
ProgenitorIs lost
Progenitorreturns
Progenitordeclares
act. as copy.
Time
Progenitor has made copies
54
Copydisconnected
Time
Copydeclares
activeCopiescreated
Replacementcreated
Copy
conn
ecte
d
Excess copies Excess deleted
A copy is disconnected from the family
55
Copydisconnected
Time
Copydeclares
activeCopiescreated
Replacementcreated
Copy
conn
ecte
d
Excess copies Excess deleted
Two active maintainers make copies
56
Copydisconnected
Time
Copydeclares
activeCopiescreated
Replacementcreated
Copy
conn
ecte
d
Excess copies Excess deleted
Disconnected copy is reconnected to the progenitor
57
Copydisconnected
Time
Copydeclares
activeCopiescreated
Replacementcreated
Copy
conn
ecte
d
Excess copies Excess deleted
Family has too many copies
58
Copydisconnected
Time
Copydeclares
activeCopiescreated
Replacementcreated
Copy
conn
ecte
d
Excess copies Excess deleted
• Copy management policies– Active: explicit removal– Passive: “natural attrition”
• Equivalent of Reynolds’ velocity matching, making and monitoring copies
USW copying policies
• Least aggressive – one at a time to chard
• Moderately aggressive – as quickly as possible to csoft and then one at a time chard
• Most aggressive – as quickly as possible to chard
• Different results59
WOs preservation status Hosts utilization statusNone
< Csoft
Csoft <= N < Chard
N == Chard
0%
< 25% < 75%
< 50 % > 75%
Dissertation, section 6.7.4
60
Least aggressive (t = 1)
61
Least aggressive (t = 10)
62
Least aggressive (t = 50)
63
Least aggressive (t = 100)
A full YouTube video is available at: http://youtu.be/sHJGYphqtK4
64
Least aggressive (final)• Results
– System stabilized
– Host capacity limited
– Some WOs without any copies
– Some hosts unused
• “Least aggressive” is not an effective policy
65
Which policy to choose?• Moderately aggressive results
in an additional 18% of WOs meeting their preservation goals and makes more efficient use of limited host resources sooner
• Most aggressive results in almost the same percentage of WOs meeting their goals, but places a strain on the host resources
Charles L. Cartledge and Michael L. Nelson, When Should I Make Preservation Copies of Myself?, arXiv preprint arXiv:1202.4185 (2012).
66
Make new family members on new hosts
• Spreading copies across hosts increases the WO’s preservation likelihood
• Learn about new hosts from friends
Dissertation, Appendix A
Reynolds’ flock
centering
Move with friends to new hosts
67
Crowd sourcing of family member creation
• “Everyone is a curator …”– Crowd sourced activity– Unscheduled– Willing to wait a long time
• Enlist humans in creation and maintenance – opposite of benign neglect
Frank McCown, Michael L. Nelson, and Herbert Van de Sompel, Everyone is a Curator: Human-Assisted Preservation for ORE Aggregations, Proceedings of the DigCCurr 2009 (2009).
68
USW simulation vs. implementation
USW Theory HTTP/HTML reality
Communications Instantaneous Asynchronous
Edges Bidirectional Directional
Temporal effects None Inconsistences
69
Some WO reference implementation details
Sawood Alam, HTTP Mailbox - Asynchronous RESTful Communication, Master's thesis, Old Dominion University, Norfolk, VA, 2013.
Carl Lagoze, Herbert Van de Sompel, Pete Johnston, Michael Nelson, Robert Sanderson, and Simeon Warner, ORE User Guide - Resource Map Implementation in Atom, Tech. report, Open Archives Initiative, 2004.
Sawood Alam, Charles L. Cartledge, and Michael L. Nelson, Support for Various HTTP Methods on the Web, Tech. Report arXiv:1405.2330 (2014).
WO memory: simulated via “edit” service
Direct WO to WO communication: simulated via the HTTP Mailbox
70
Demonstration of the reference implementation
1. Selection of a web page to be preserved2. Creation of a WO from the web page3. Adding the WO to an existing USW graph
a. Pages copied from flickr.com, arXiv.org, radiolab.org, and gutenburg.org
b. All pages instrumented to become USW WOs
4. Creating preservation copies5. Detecting that a copy was lost6. Creating a replacement copy
71
USW contributions
71
Expanded graph theory by creating an algorithm that creates small-world graphs based on locally collected data (chapter 6)
Developed a new way to quantify damage in connected and disconnected graphs (section 5.2)
Developed techniques to optimize when and where to create preservation copies (section 5.5)
Developed techniques to achieve emergent behavior in WOs (section 6.2)
72
Backup slides
73
Preserve Me Viz! with new connections
• New friend connections
• New copy locations
74
Preserve Me “Basic” on a copy• Differences between
active and passive maintainers.
• Active maintainer is responsible for making copies.
• Passive maintainer sends alerts to the active maintainer
• Passive maintainer may assume active maintainer role if active is not available.
75
A USW instrumented splash page
…<link rel="resourcemap" type="application/atom+xml;type=entry" href="http://arxiv.cs.odu.edu/rems/arxiv-0704-3647v1.xml" /><link rel="aggregation" href="http://arxiv.cs.odu.edu/rems/arxiv-0704-3647v1.xml#aggregation" /><script src="http://www.cs.odu.edu/~salam/wsdl/uswdo/work/preserveme.js"></script>…
76
USW algorithm popup• Written in
JavaScript• Relies on domain
services– Copy -> creates
copy of a WO– Edit -> update own
REM• Uses
communications mechanism based on Sawood Alam’s master’s thesis
77
USW Preservation: copies (1 of 2)
• WO copies are not bit by bit identical to the original WO
• REsource Map (REM) points to a resource– Point to the “essence” of the
original– Point to local copies of the
resources– Can be used to recreate the
“essence” of the original• Resource has two attributes:
– Size– Update frequency
78
USW Preservation: copies (2 of 2)
79
Watts Stogratz small-world growth
80
Graph theory: random graph
81
Graph theory: scale free graph
82
Graph theory: lattice graph
83
Graph theory: Watts and Strogatz small-world graph
84
Comparing graphs
• Small-world graphs occur in natural and man made systems
• Small-world graphs are robust• How to algorithmically and
incrementally create small-world graphs?
Symbol Meaning
K Degree
<k> Average degree
N Order of the graph
85
Quantifying damage
• All graph components are not equally valuable
• How to identify most valuable
• Greedy repair is the obverse of identifying the most damaging component by identifying where to place the most beneficial component
Charles L. Cartledge and Michael L. Nelson, Connectivity Damage to a Graph by the Removal of an Edge or Vertex, Tech. report, arXiv 1103.3075, ODU CS Dept.
86
Global normalized A{DV}H damage (40 - 750 nodes)
• Arithmetic series of possible solutions
• Early attacks are most effective, later attacks are incrementally effective
87
Long term growth analysis of USW graph
Based on the idea of a game
– Create the graph– Attack the graph using AVH
profile to remove 10% of the WOs
– Repair the graph, every surviving WO gets 2 opportunities (may be unsuccessful in repair attempts)
– Repeat until steady state
88
Possible paths in attack/repair game
Dissertation section 6.9
89
USW copies: famine to feast
90
Final states for copying policies and named conditions
Dissertation, Appendix H
91
Host capacity and WO desires
Famine FeastStraddle
B. L
ow
B. H
igh
Dissertation, Appendix H
92
Man-made small-world graph: Western Electricity Coordinating Council
Western Electricity Coordinating Council
Actual Random graph
Nodes Edges C(G) L(G) C(G) L(G)4941 6594 0.0801 18.99 0.00054 8.7
Ake J Holmgren, Using Graph Models to Analyze the Vulnerability of Electric Power Networks, Risk Analysis 26 (2006), no. 4, 955 - 969.
93
Naturally occurring small-world graph: C. elegans nematode
Caenorhabditis elegans
Actual Random graph
Nodes Edges C(G) L(G) C(G) L(G)248 511 0.21 2.87 0.05 2.62
Lav R Varshney, Beth L Chen, Eric Paniagua, David H Hall, and Dmitri B Chklovskii, Structural Properties of the Caenorhabditis elegans Neuronal Network, PLoS computational biology 7 (2011), no. 2, e1001066.
94
Organic small-world graph: Enron e-mail
Enron e-mail
Actual Random graph
Nodes Edges C(G) L(G) C(G) L(G)148 ~500,000 0.44 2.25 0.11 2.0
Shinako Matsuyama and Takao Terano, Analyzing the ENRON Communication Network Using Agent-Based Simulation, Journal of Networks 3 (2008), no. 7.
95
USW WO determines number of friends
• Number of connections
Dissertation, section 6.7
Symbol Meaning
ln, log2 Natural and base 2 logarithms
n Order of the discovered USW graph
g Simple scalar
96
Wandering activities
Dissertation, Appendix B
97
Active maintenance activities
Dissertation, Appendix B
98
Passive maintenance activities
Dissertation, Appendix B
99
Video URLs• USW video
– http://youtu.be/JnCMenp73YQ• Least Aggressive
– http://youtu.be/sHJGYphqtK4• Moderately Aggressive
– https://www.youtube.com/watch?v=pVI-VhPh7KQ• Most Aggressive
– https://www.youtube.com/watch?v=eIXz8Njh-QM• “Death Star” message histogram
– https://www.youtube.com/watch?v=X3EShyjFoc4• “Traditional” message histogram
– https://www.youtube.com/watch?v=9CcCup3Td-Q
100
Useful URIs• Flickr
– https://www.flickr.com/• Flickr on cs.odu.edu
– http://flickr.cs.odu.edu/• Adding image
– http://ws-dl-02.cs.odu.edu:10102/rem/generate/0.8/0.95/• Preseve Me! Viz
– http://www.cs.odu.edu/~salam/preserveme/viz.html• Delete REM
– http://ws-dl-02.cs.odu.edu:10102/rem/remove/http://flickr.cs.odu.edu/rems/flickr-24791103-N07-6661587389.xml
• Sawood on flickr– https://www.flickr.com/photos/122128913@N05
• Chuck on flickr– https://www.flickr.com/photos/24791103@N07/
• Court de Tomas De Torquemada on flickr– https://www.flickr.com/photos/24791103@N07/12867674403/