Presentation is loading. Please wait.

Presentation is loading. Please wait.

High Performance Simulation of Internet Worms David M. Nicol Dept. of ECE, & Information Trust Institute University of Illinois, Urbana-Champaign.

Similar presentations


Presentation on theme: "High Performance Simulation of Internet Worms David M. Nicol Dept. of ECE, & Information Trust Institute University of Illinois, Urbana-Champaign."— Presentation transcript:

1 High Performance Simulation of Internet Worms David M. Nicol Dept. of ECE, & Information Trust Institute University of Illinois, Urbana-Champaign

2 2 problem background Worm: malware that through automated means gains access to a host, implants a copy of itself, and uses the host to attempt to infect others Evaluation of defense against worms most readily facilitated by simulation / emulation –e.g. use simulator to generate traffic thrown against defenses Worms that search for targets can generate a lot of traffic –fast-scanning worms saturate links and may topple routers Problems : Find efficient and accurate ways of simulating infection growth of random scanning worms Find efficient and accurate ways of simulating scan traffic across large networks

3 3 variance is important Random scanning induces variation in infection times. It can be significant 20 independent sample paths, Code Red vII parameters 10% of infections Variance in time of k-th infection Variance in I(t) Early life : when the worm must be detected if defenses are to be effective Our evaluations need to be based on a range of possible behaviors one sample path is not enough

4 4 detailed model Like Code Red vII (except for uniform scanning) –100 concurrent infected host –Long time-out delay when SYN is not acked –Short delay with SYN-ACK return –State-independent probability of receiving SYN-ACK –State-dependent probability of infection from responding host TCP open timeout SYN-ACK returned,no infectioninfection Inner cycle Outer cycle Very small probabilities of infection imply geometric # inner cycles can be replaced with an exponential distribution

5 5 Time of Next Infection (TNI) Given I current infected hosts, time to next infection is the minimum of 100xI independent exponentials, hence is exponential with rate Scan rate Number of targetsProb{hit a target} State evolution of I(t) is thus a continuous-time Markov chain Compare this formulation with the classic SI equations TNI is a discrete version of SI

6 6 TNI and variance Sum of inter-arrival times Grows small quickly with increasing j Variance is dominated by early inter-infection times

7 7 TNI verification Time of ith infection, and I(t) are essentially indistinguishable at 95% significance level (K-S test)

8 8 hybrid model TNI is more efficient when susceptibles are hard to find, SI is more efficient when I(t) is rising rapidly Early life : few infections mid-life: many infections, many susceptibles Late life: few susceptibles  Decision policy : infection to simulate with TNI, or with numerical integration of SI Problem : minimize execution time, subject to constraint that relative error in variance for every time of infection is bounded

9 9 results a) For any  and infection index i, in logarithmic time find largest index f(i,  ) s.t. relative error in variance is <  if TNI used before i and after f(i,  ) Infection number i f(i,  ) TNI

10 10 Infection number Numerical integration TNI sampling results b) Cost of numerical integration of SI as a function of infection index is convex and symmetric, while cost of TNI per infection is constant c) For given i and f(i,  ) we can compute the least cost policy d) Optimum policy is tractable Numerical integration i f(I,  )

11 11 ratio of SI solution cost to hybrid cost Size of each subnetwork relative to Internet Factor by which S(0) exceeds avg. per subnet Very high ratio Approx. 10% of subnet hosts are susceptible

12 12 ratio of TNI cost to hybrid cost Size of each subnetwork relative to Internet Factor by which S(0) exceeds avg. per subnet Densely distributed susceptibles While TNI is “usually” much faster than differential equation model, hybrid protects against the dense cases

13 13 backbone and sub-networks Imagine a large backbone network –Routers modeled in detail –Sub-networks attach through backbone routers Worm state advanced individually in sub-networks Each  time, each sub-network offers description of scan flows (dest,rate) to backbone simulator Problem now is to determine delivered scan rates to each sub- network AT&T AboveNet Exodus Cable&Wireless Level3 Verio Sprint UUNet

14 14 Resolution and Transparency All of a port’s final output flows can be resolved once all of its input flow values are resolved But to break cycles we need to be smarter…. A port is transparent if the sum of input rate bounds is no greater than the output bandwidth Example : Suppose Then 1. so that Port becomes transparent Try to resolve final output flow values based on upper bounds Notice that every output flow is bounded from above by input flow rate …. Every flow can be bounded by its ingress rate Flow rate becomes resolved 2. Port becomes resolved Port becomes resolved 3. Flows become resolved 4. Repeat

15 15 Modeling congestion Even though flows are acyclic, dependency cycles may form in definition of flow rates congestion when otherwise Define in out in depends on No congestion out

16 16 State Change Rules Port states are {resolved, transparent, unresolved} Flow states are {settled, bounded} Rule 1: port resolution Pre-condition Action Port state is not resolved and all input flow states are settled Mark port state as resolved, compute all output flow values, mark each as settled

17 17 State Change Rules Port states are {resolved, transparent, unresolved} Flow states are {settled, bounded} Rule 2: port transparency Pre-condition Action Port state is unresolved and sum of input rate bounds is less than bandwidth, Mark port state as transparent. For every input rate that is settled, mark corresponding output rate as settled

18 18 State Change Rules Port states are {resolved, transparent, unresolved} Flow states are {settled, bounded} Rule 3: settle state transition Pre-condition Action Port state is transparent, some input flow is settled, and corresponding output flow is not Mark corresponding output flow as settled, with value equal to input flow value

19 19 State Change Rules Port states are {resolved, transparent, unresolved} Flow states are {settled, bounded} Rule 4: flow bound transition #1 Pre-condition Action Port state is unresolved, the fair proportion relative to settled flows of an input flow rate exceeds bound on output flow Lower corresponding output flow bound to be equal to fair proportion of input flow bound is sum of settled flow rates

20 20 State Change Rules Port states are {resolved, transparent, unresolved} Flow states are {settled, bounded} Rule 5: flow bound transition #2 Pre-condition Action Port state is not resolved, the flow rate bound of an input flow is less than the corresponding output flow bound Set bound of output flow equal to bound on input rate

21 21 Cycle Resolution After all that, we may still be left with cycles of unresolved ports General problem is solution of a system of non-linear equations –Solution methods generally iterative The number of iterations, and cost of iterations is principle issue –We explore “fixed-point” iteration. Each iteration : –freeze all input rates –compute output rates based on frozen input rates –compare new solutions with old for convergence Our experiments define convergence when the relative difference between successive flow value solutions is less than (1/10)% for all flow values

22 22 Results Convergence behavior –Examine # ports in cycle and iterations for convergence –Vary topology Topology#routers#links#flowsMbps Top Top Top Top Topologymedian #ports in cycles#median iterations Top Top Top Dependency reduction is effective Fixed point algorithm converges quickly

23 23 Results We ask –How fast does it run? –What is speedup relative to pure packet simulation? Topology#routers#links#flowsMbps Top Top Top Top Topologysecs/time-stepsecs/time-step (20% link util.) (50% link util.) Top Top Top Top For 1 sec time-step, faster than real- time on a model equivalent to 1.9G pkt-evts/sec (1K bytes/pkt) Experiments run on PC 1.5 GHz CPU 3Gb memory Linux OS Topologies Results

24 24 Results We ask –How fast does it run? –What is speedup relative to pure packet simulation? Topology#routers#links#flowsMbps Top Top Top Top Experiments run on PC 1.5 GHz CPU 3Gb memory Linux OS Topologies Results Link util.speedupLink util.speedup 10% % % % % % % % 1135 Directly compare packet-oriented simulation, using exactly same input flow rates, on Top-1 speedup over wide range of loads

25 25 Primer on packet forwarding CIDR addressing : /24 Forwarding decisions represented in trie –Branch on most significant bits in IP address –Remember last node seen with interface label –Example : , b,3 a,2 c,1 e,4 d, labelIP spaceOutput interface a00*2 b0100*3 c1*1 d1000*5 e101*4

26 26 flow optimizations Number of unique source-destination scan flows grows in the square of the number of sub-networks Run-time can be improved by reducing # flows Exploit common situation that payload does not matter for modeling –Flows are sufficiently identified by (dest,rate) –We can merge flows targeting the same destination, just add rates ( , 5) ( ,64, 27) ( , 19) No merging ( , 5) ( , 27) ( , 19) ( , 46) Merging

27 27 merge improvement Reduction of # flows is topology dependent For illustrative purposes consider a tree Count “flow crossings” : sum over all links of # flows Common destination Flow compression ratio is approximately log(# subnetworks )/ 2

28 28 splitting flows During peak worm scanning a sub-network will be scanning all other sub-networks Reduce # flows by offering to network an aggregate description, e.g /0 (all IP addresses) A router must split an aggregate flow descriptor when subspaces are routed through different interfaces ( /8,400) ( /10,100) ( /10, 100) ( /9, 200)

29 29 splitting and tries Push labels of exit points up the trie b,3 a,2 c,1 e,4 d, sink:3 sink:2 sink:3 sink:5 sink:4 labelIP spaceOutput interface a00*2 b0100*3 c1*1 d1000*5 e101*4 Splitting Algorithm a)Follow MSB of prefix to completion, or node with conflicted children b)If completed, route all to implied interface. If conflicted children Build set of descendents with earliest sink labels Partition flow in proportion to routed space

30 30 run-time gains are good Very fast worm, 4 backbone networks spanning O(100) to O(1000) routers : merge+split yields x15 improvement in execution time

31 31 memory gains are good Very fast worm, 4 backbone networks spanning O(100) to O(1000) routers : merge+split yields x10 reduction in flow interfaces

32 32 summary Developed hybrid model of worm whose execution time is optimal subject to constraint on error in variance Consider backbone + sub-networks model –Developed fast solution for solving huge system of coupled non-linear equations describing bandwidth sharing –Developed optimizations that reduce # flows managed in backbone by an order of magnitude We are simulating the interactions of infrastructure and worms on very large examples


Download ppt "High Performance Simulation of Internet Worms David M. Nicol Dept. of ECE, & Information Trust Institute University of Illinois, Urbana-Champaign."

Similar presentations


Ads by Google