agilis: an on-line map reduce environment for collaborative security

Sapienza Università di RomaDipartimento di Informatica e Sistemistica

Middleware LaboratoryMIDLAB

AGILIS: an on-line map reduce environment for collaborative security

Roberto Baldoni

Università degli Studi di Roma “La Sapienza”

[email protected], http://www.dis.uniroma1.it/~baldoni/

Prin Meeting - San Vito di Cadore

Joint Work with IBM Haifa

in the context of CoMiFin EU Project

14/2/2011

mailto:[email protected]

http://www.dis.uniroma1.it/~baldoni/

http://www.dis.uniroma1.it/~baldoni/

Mid

dle

ware

Labora

tory

MID

LAB

Focus and structure of the talk

■ Requirements coming from the financial context;

■ Collaborative event processing for Cyber Security

■Edge vs centralized event processing over the internet

■ Agilis

■Esper

Roberto Baldoni



The case of the Financial Critical Infrastructure

Mid

dle

ware

Labora

tory

MID

LAB

The case of Collaborative Cyber Security in Financial Ecosystem

■"webification" of critical financial services, such as home banking, online trading, remote payments;

■Cross-domain interactions, spanning different organization boundaries are in place in financial contexts;

■Heterogeneous infrastructure systems such as telecommunication supply, banking, and credit card companies working on heterogeneous data;

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

The case of Collaborative Cyber Security in Financial Ecosystem■ A payment card fraud (2008)

■100 compromised payment cards used by a network of coordinated attackers retrieving cash from 130 different ATMs in 49 countries worldwide, totaling 9 million of US dollars.

■High degree of coordination, half an hour to be executed

■evade all the local monitoring techniques used for detecting anomalies in payment card usage patterns.

■The fraud has been detected only later, after aggregating all the information gathered locally by each financial institution involved in the payment card scam

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

The case of Collaborative Cyber Security in Financial Ecosystem

■Distributed Denial Of Service Attack (2007, Northern Europe)

■ render web-based financial services unreachable from legitimate users.

■DDoS attack targeted a credit card company and two DNS.

■Internet restored only after several trial-and-error activities carried out manually by network administrators of the attacked systems and of their Internet Service Providers (ISPs).

■Long preparation time (days), short attack time (seconds)

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Economics of a DDOS

■render internet-based financial services unreachable from legitimate users.

■Use of Botnets (rented now with a credit card in a few minutes)

■Three examples of DDOS campaign in Cyberwarfare:

■Estonia 2007

■Georgia 2008

■Iran (in progress!). Stuxnet worm invaded Iran’s Supervisory Control and Data Acquisition systems

McAfee report 2010 “in the crossfire: critical infrastructures in the age of cyber war “

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Economics of a DDOS

McAfee report 2010 “in the crossfire: critical infrastructures in the age of cyber war “

■cost of downtime from major attacks exceeds U.S. $6 million per day

■damage to reputation

■loss of personal information about customers

■one out of five DDos attacks is accompanied with an extorsion

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

The case of Collaborative Cyber Security in Financial Ecosystem■Both previous attacks cannot be detected quickly

through information available at the IT infrastructure of a single financial player (i.e., using local monitoring)

■Need of Information Sharing

■Exchange non-sensitive status information

■Set up of agreements

■Advantages of a global monitoring system

■Damage mitigation

■Quick reaction

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Barriers to Collaboration

Internet

LLYODS

Unicredit

France Telecom

EDF

AT&T

SWIFT

Events

warnings

UBS

■Barriers to collaboration

■Understanding the economics

■Trust

■Legal Issues

Roberto Baldoni



Collaborative event Processing for cyber security:

The CoMiFin Project

Internet level

Collaboration Level

Application Level

Mid

dle

ware

Labora

tory

MID

LAB

Collaborative Cyber Security Platform

■Monitoring and reaction to threats (MitM, Stealty Scan , Phishing, …)

■Black/white lists distribution (for credit reputation, trust level, …)

■Anti-terrorism lists (with name check VAS)

■Anti money laundering monitoring

■Risk management support

■ Some Requirements on the platform

■ uneven workload along the time

■ High throughput

■ high computational power

■ Large storage capabilities

■ Timeliness

■ Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

The notion of semantic room

■Contract

■ set of processing and data sharing services provided by the SR along with the data protection, privacy, isolation, trust, security, dependability, performance requirements.■ The contract also contains the hardware and software requirements a member has to provision in order to be admitted into the SR.

■ Objective

■ each SR has a specic strategic objective to meet (e.g, large-scale stealthy scans detection, detecting Man-In-The-Middle attacks)

■ Deployment

■ highly flexible to accommodate the use of different technologies for the implementation of the processing and sharing within the SR (i.e., the implementation of the SR logic or functionality).

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

The notion of semantic room: relationship with cloud computing

Internet

Level

Collaboration

Level

Application

Level

■Private cloud

■ Deployment of the semantic room through the federation of computing and storage capabilities at each member

■ Each member brings a private cloud to federate

■Public Cloud

■ Deployment of the semantic room on a third party cloud provider

■ The third party owns all computing and storage capabilities

■Hybrid approach

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Data Management problems in the semantic room

■ Jurisdiction and regulation (Where and how will data be governed?)

■ Ownership of Data (Who owns the data in the semantic room?)

■ Data Portability

■Data anonymization

■Data Retention/Permanence (What happens to data over time?)

■Security and Privacy (How is data secure and protected?)

■Reliability, Liability and Quality of Service of the partner of the semantic room

■ Government Surveillance (How much data can the government get from a semantic room?)

■………………….Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

A specific collaborative platform: CoMiFin Architecture

contract

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Related work

Roberto Baldoni

■ IBM System S [ICDCS 06]

■ high cost of ownership

■Centralized data management

■No cooperative approach

■Cooperative Intrusion Detection Systems (e.g. Dshiels)

■Correlation among local warnings

■ High cost of ownership

■ Obscure data management



Preventing Stealthy Scan Through centralized processing

Internet level

Collaboration Level

Application Level

Mid

dle

ware

Labora

tory

MID

LAB

Collaborative Stealthy scan

Attacker performs port scanning simultaneously at multiple sites trying to identify TCP/UDP ports that have been left open. Those ports can then be used as the attack vectors

Added value of collaboration:

■Ability to identify an attacker trying to conceal his/her activity by accessing only a small number of ports within each individual domain

Action taken:

■black list IP addresses

■update historical records

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Collaborative Stealthy scan detection

Attack subjects:

■External web servers in DMZ’s of the SR members

Pattern:

■“Unusually” high number of requests

■Originating from a particular source IP address, and

■Directed to distinct (machine, port) pairs

Action taken:

■Matching source IP’s are banned from the future access to external web servers

Defining the attack

■Use of common scanning tools (nmap)

■Use of real trace (e.g., ITOC US Army)

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Collaborative Stealthy scan detection

■Rank-Syn algorithm.

■ Analyze the sequence of SYN, ACK, RST packets in the three-way TCP handshake. Specifically, in normal activities the following sequence is verified (i) SYN, (ii) SYN-ACK, (iii) ACK.

■ In the presence of a SYN port scan, the connection looks like the following: (i) SYN, (ii) SYN-ACK, (iii) RST (or nothing)

■ For a given IP address, if the number of incomplete connections is higher than a certain threshold T, we can conclude that the IP address is likely carrying out malicious port scanning activities.

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Stealthy scan detection

Raw data: TCPdump10:53:14.647181 IP 9.148.30.136.pop3 > 9.148.17.85.madcap: R 0:0(0) ack 1 win 0

10:53:14.653813 IP 9.148.17.85.sis-emt > 9.148.30.136.xfer: S 268426387:268426387(0) win 65535 <mss 1460,nop,nop,sackOK>

10:53:14.653817 IP 9.148.30.136.xfer > 9.148.17.85.sis-emt: R 0:0(0) ack 268426388 win 0

Normalized data format:

■LogEvent: sourceIp, destIp,sourcePort, destPort, startTime, endTime, bytesSent, bytesRecieved, returnStatus;

Online data summary

Historical Data Format

BlackList: List of IP addresses

Source IP

Rate

Source IP

Requests count

Port count

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Example of semantic room for stealthy scan: Ingredients

Roberto Baldoni

Branch jBranch 1Gateway

I/O socket

adapter

sniffer

Branch NGateway

I/O socket

adapter

sniffer

.

.

.

POJOs

POJO

s

Esper CEP Engine

I/O socket

Main Engine

Input Streams

Output Streams

Scanner list

suspected IPs

EPL QueryEPL Query

EPL Query

EPL QueryEPL Query

Subscriber

Mid

dle

ware

Labora

tory

MID

LAB

Example of semantic room for stealthy scan: Ingredients

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Testbed: latency measurement

Two Semantic rooms

■Man-In-The-Middle Attack

■Stealthy Scan

Roberto Baldoni



Preventing Stealthy Scan through edge processing

Internet level

Collaboration Level

Application Level

Mid

dle

ware

Labora

tory

MID

LAB

Example of semantic room for stealthy scan: Ingredients■WebSphere eXtreme Scale (WXS): in-memory

distributed storage

■High-level language for processing logic: Jaql (SQL-like, supports flows)

■Distributed processing runtime: MapReduce

■Distributed file system for long-term storage: HDFS

■ Agilis consists of a distributed network of processing and storage elements hosted on a cluster of machines (also geographycally dispersed)

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Data Dissemination: Agilis

Roberto Baldoni

Gateway

JaqlInterpr

eter

Jaql query

AGILIS

Re-define InputFormat,

OutputFormat

Distributed In-Memory Store (WXS)

Storage container

Storage container Cat 1

Cat 2

Distributed File System (HDFS)

Jaql Adapter

Map-Reduce (Hadoop)

Job Tracker

TaskTracker

TaskTracker

TaskTracker

HDFS Adapter

WXS Adapter

Mid

dle

ware

Labora

tory

MID

LAB

Example of semantic room for stealthy scan: architecture

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Collaborative Stealthy scan detection with Agilis

Detection of stealty scan

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Demo: Done at Haifa IBM Research LAB (2009)

■Simple and homemade attacks

■ artificial traces

■Simple stealty scan detection algorithm

8 Linux Machines on a LAN, each of which with

■2GB of RAM and

■20GB of disk space

■One machine was hosting all the management processes (JT, XS Catalogue)

Each of the remaining 7 hosts modeled a single SR participant

■DMZ web server under attack

■TT and XS data server

Scenarios:

■Single intruding host that generated a series of TCP/SYN requests targeting a fixed set of 300 unique ports on each the 7 attacked servers

■requests injected at constant rate of 10, 20, and 30 req/server/sec

■ratio of attack to legitimate traffic 1:5

■blacklisting threshold: 20,000 requests and 1000 unique port

■processing window: 4 minutes

Results:■No overload

■Detection latency 700 sec, 430 sec, 330 sec

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Demo: work we are doing in our LAB (2010)

The video shows a semantic room implemented in Agilis using

■Nmap for producing the attack

■Real TCP dumpsJoint work with Giorgia Lodi and Leonardo Aniello

Roberto Baldoni

Mid

dle

ware

Labora

tory

MID

LAB

Testbed: latency measurement

Two Semantic rooms

■Man-In-The-Middle Attack

■Stealthy Scan

Roberto Baldoni