Modular RADAR: Immune System Inspired Strategies for Distributed Systems Soumya Banerjee and Melanie Moses University of New Mexico
May 11, 2015
Modular RADAR: Immune System Inspired Strategies for Distributed
Systems
Soumya Banerjee and Melanie MosesUniversity of New Mexico
Outline• Distributed systems and the natural immune system (NIS)
operate under similar constraints• Effect of body size on NIS search and response times• Scale invariant detection and response• Hypothesis: architecture of the lymphatic system leads to
invariant search and response times• Modular RADAR strategy• Number and size of lymph nodes increases with organism
size• Distributed systems
– P2P system– Multi-robot control
• Future directions
Properties of Distributed Systems
• Physical space is important• Resource constrained (power, bandwidth)• Performance scalability is a desirable feature
• Operates under constraints of physical space• Resource constrained (metabolic input,
number of immune system cells)• Performance scalability is an important
concern (mice to horses)
Properties of the Natural Immune System (NIS)
Problems Faced by the NIS• Only a few NIS
cells are specific to a particular pathogen ( in
T-cells)
€
106
€
1
Search Problem
• They have to search throughout the whole body to locate small quantities of pathogens
Response Problem
• Have to respond by producing antibodies
West Nile Virus infection 25 species of birds and 4 species of mammals infected with WNV
• Bunning et al. (2002)• Komar et al. (2002)
Unimodal peak at ~ 2 to 4 days post infection
Immune response rates and times are not correlated with host mass
• assuming immune response causes peak• B-cell response in mice ~ 4 days
Komar et al. 2002
• Experimental data indicates that the NIS can search for pathogens and respond by producing antibodies in time invariant of organism body size
Nearly Scale-Invariant Search and Response
Nearly Scale-Invariant Search and Response
• How does the immune system search and respond in almost the same time irrespective of the size of the search space?
Solution: Lymph Nodes (LN)• A place in which IS cells and the pathogen can
encounter each other in a small volume• Form a decentralized detection network
Crivellato et al. 2004
Modular RADAR
• Search is now– modular– efficient– parallel
We call this a modular RADAR (Robust Adaptive Decentralized search Automated Response)
Hypothesis• Architecture of the immune system is
responsible for nearly scale-invariant search and response properties
• We now focus on West Nile Virus
www.lymphadvice.com
Lymph Node Dynamics
Lymph Node Dynamics
Lymph Node Dynamics
€
T = tdetectDC + tmigrate
DC + tdetectcTcell ,DC + trecruit
Scaling of LN Size and Number
• this is in qualitative agreement with data• need more data
€
T = tlocal + tglobal
T = tdetectDC + tmigrate
DC + tdetectDC ,cTcell + trecruit
After minimizing we have
N ∝M 4 / 7,where N is the number of LNs
VLN ∝M3 / 7,where VLN is the size of a LN
Banerjee and Moses 2010, Swarm Intelligence (under review)
Modular RADAR Architecture
€
T = tlocal + tglobal ∝M1/ 7
Summary
• There are increasing costs to global communication as organisms grow bigger
• Semi-modular architecture balances the opposing goals of detecting pathogen (local communication) and recruiting IS cells (global communication)
• This leads to scale invariant detection and response
• Can we emulate this modular RADAR strategy in distributed systems?
Peer-to-Peer Systems
• Used to provide distributed services like search, content integration and administration
• Computer nodes store data or service • No single node has complete global
information • Decentralized search using local information
to locate data
Semantic Small World (SSW) P2P Overlay Network
• Represents objects by a collection of attribute values derived from object content
• Aggregates data objects with similar semantics close to each other in clusters in order to facilitate efficient search
• It maintains short and long-distance connections between clusters.
• The long-distance connections follow a precise probability distribution making the whole overlay network small-world (Kleinberg 2000)
* M. Li et al. 2004
Semantic Small World (SSW) P2P Overlay Network
adapted from M. Li et al. 2004
Bounds for Efficient Decentralized Search in SSW
• Average search path length for search across clusters is
where n is the total number of nodes, c is the number of nodes in a cluster,
l is the number of long-distance connections per node
€
tglobal =Olog2(n /c)
l
⎛
⎝ ⎜
⎞
⎠ ⎟
M. Li et al. 2004
SSW with Modular RADAR
• Our contribution is to – vary the cluster size– vary the number of long-distance connections
as
– such densification is seen as an emergent property of technological networks (Kleinberg 2004) and also incorporates redundant paths€
l = log(n /c) = log(numclusters)
tglobal =O(log(n /c))
Time to Search in SSW with Modular RADAR
minimizing by differentiating with respect to c we have
€
T = tlocal + tglobal
T =α 1c1/ 2 +α 2 log(n /c)
€
c =O(log2 n)
T =O(logn − loglogn)
SSW with Modular RADAR
Wireless Mobile Devices: Original System
adapted from Nair et al. 2008
Tradeoffs• Potential communication bottlenecks
– local communication between robots and computer servers – global communication between computer servers
• If both local and global communication are constrained, then sub-modular architecture balancestradeoff
System modified with modular RADAR
Future Directions• Strategy is widely applicable• A modular RADAR strategy can be used to augment
– Intrusion Detection Systems (Hofmeyr and Forrest 1999)
– Multi-Robot Control– Wireless Sensor Networks– Wireless Devices (Specknets: Hart and Davoudani
2009)– Collective Robotic Systems using Artificial Lymph
Node Architectures (Mokhtar, Timmis, Tyrrell and Bi 2008)
Summary• The NIS and distributed systems operate under similar
constraints• Physical space of organism body constrains NIS search and
response times• The NIS has evolved a sub-modular RADAR architecture in
which LN numbers and sizes increase with organism body size
• This balances the tradeoff between local communication (pathogen detection) and global communication (antibody production); this leads to scale invariant detection and response
• Similar tradeoffs also exist in distributed systems• Such a modular RADAR approach is shown to improve
search times in P2P and multi-robot control systems• Can be applied in other distributed systems
Acknowledgements
• Dr. Melanie Moses• Dr. Alan Perelson• Dr. Stephanie
Forrest• Dr. Jedidiah Crandall• Dr. Rob Miller• Dr. Sam Loker
• SFI Complex Systems Summer School
• Travel grants from PIBBS (Dept. of Biology, UNM)
• Travel grants from RPT and SCAP (UNM)
• NIH COBRE CETI grant (RR018754)