Dynamic Quarantine of Internet Worms

Dynamic Quarantine of Internet Worms

The International Conference on Dependable Systems and Networks (DSN-2004). Palazzo dei Congressi, Florence, Italy. June 28th - July 1, 2004. Dynamic Quarantine of Internet Worms Cynthia Wong, Chenxi Wang∗, Dawn Song, Stan Bielski, Gregory R. Ganger† Carnegie Mellon University {cindywon, chenxi, dawnsong, bielski, ganger}@cmu.edu Abstract from host to host, restricting the contact rate of a worm con- strains how fast the infection can spread in the network. If we limit the contact rate of worm traffic, can we al- Previous proposals of rate control consider deploying such leviate and ultimately contain Internet worms? This paper mechanisms primarily at the individual host level. In this sets out to answer this question. Specifically, we are inter- paper, we investigate rate control at individual end hosts and ested in analyzing different deployment strategies of rate at the edge and backbone routers, for both random propaga- control mechanisms and the effect thereof on suppressing tion and local-preferential connection worms. Our analysis the spread of worm code. We use both analytical models shows that both host and edge-router based rate control re- and simulation experiments. We find that rate control at in- sult in a slowdown (in the spreading rate of the worm) that dividual hosts or edge routers yields a slowdown that is lin- is linear to the number of hosts (routers) implementing the ear in the number of hosts (or routers) with the rate lim- rate limiting filter. In particular, host-based rate control has iting filters. Limiting contact rate at the backbone routers, very little benefit unless rate limiting filters are universally however, is substantially more effective—it renders a slow- deployed. Rate control at the backbone routers, however, is down comparable to deploying rate limiting filters at every substantially more effective. Our results hold true for both individual host that is covered. This result holds true even random propagation worms (e.g., Code Red I) and worms when susceptible and infected hosts are patched and immu- that spread via a preferential connection algorithm such as nized dynamically. To provide context for our analysis, we those that target local hosts within a subnet. examine real traffic traces obtained from a campus comput- Results are similar when dynamic immunization is taken ing network. We observe that rate throttling could be en- into account. As the worm spreads and the knowledge of forced with minimal impact on legitimate communications. the worm disseminates, an increasing number of hosts (both Two worms observed in the traces, however, would be sig- infected and susceptible) will be patched, immunized and nificantly slowed down. consequently removed from the susceptible population. In an effort to study realistic worm attacks, the models in this paper incorporate dynamically changing the immunization rates. This is in contrast to the traditional models for which 1. Introduction the rate of immunization remains constant throughout the infection outbreak [7, 16, 2, 6, 15]. Since the original “Internet worm” [3] in 1988, computer To provide context for the models, we examine traffic worms continue to wreak havoc on the Internet. The recent traces obtained from a sizable campus computing network. SQL Slammer worm infected over 90% of the vulnerable We observe that limiting the rate of unique IP addresses hosts on the Internet within ten minutes [10]. With such contacted (as in [17]) from the edge of the departmental net- voracity, the manual patch-’em-as-they-go approach simply work to no more than 16 (total contacts) per five-second pe- does not work. We need automated detection and response riod would almost never affect legitimate traffic. Individual to defend against worm outbreaks. host rates can be kept to under four per five-second period. One class of techniques that seems promising is rate Limiting only non-DNS-translated IP address contacts [5] control—schemes that aim to limit the contact rate of worm can reduce the contact rate by another factor of2–4.Our traffic [5, 17]. Since worms typically spread at a rapid speed traces also captured the behavior of machines infected by two worms: Welchia and Blaster. The results confirm that ∗ This work was partially supported by the National Science Foundation infected machines exhibit much higher contact rates and under Grant No. CCR-0208853 and ANI-0326472. could be dramatically slowed by rate limiting. † This research is partially sponsored by the Air Force Research Labo- Combining practical rate limits with our models allows ratory, under agreement number F49620-01-1-0433. us to estimate how well such approaches might work in practice. For instance, to secure an enterprise network from every individual has equal contact to every one else in the worms that propagate using a local-preferential connection population. This assumption is similar to the ways in which algorithm, our study shows that unless rate limiting filters random propagation worms spread in computer networks. are deployed at both the edge routers and a certain percent- This model is described in more detail in [1]. A homoge- neous model assumes a connected network with N nodes. It age of the end hosts, little benefit will be gained. β The remainder of this paper is organized as follows. Sec- also assumes an average infection rate across all links. If we represent total number of infected nodes at time t as I t,a tion 2 describes related work. Section 3 gives a brief back- deterministic time evolution of I (infected hosts) can be ob- ground in epidemiological models. Sections 4 and 5 study tained as below, deployment strategies of rate limiting schemes. Section 6 dI t = βI (N − I /N ) incorporates dynamic immunization with rate control, and dt t t (1) Section 7 presents a case study of real network traces. We I/N = eβt c summarize in section 8. The solution to Equation (1) is c+eβt , where is a constant. c is determined by the initial infection level. 2. Related work c → N − 1 when the initial infection level is low, since the fraction of infected hosts will be small. Several documented studies investigated computer From this we can see that the infection grows exponen- worms and the ways in which they propagate. Stani- tially initially and reaches saturation after a certain point. α ford et al. presented a study of different types of worms The time takes to reach a certain infection level is . and how they can cause damage on the Internet [13]. t = lnα/β (2) Zou et al. [19] analyzed the propagation of the Code Red worm and presented an analytic model for worm propa- The analytical models described in the later parts of this gation; Moore et al. [10] analyzed the propagation of the paper are derived from the basic homogeneous model and Slammer worm and its effect on the Internet. These stud- share the same assumptions. ies have not analyzed defense mechanisms in great depth. Moore et al. [11] explored the design space for worm 4. Rate Limiting containment systems. They studied the efficacy of address blacklisting and content filtering with various deployment In this section, we present a study on rate limiting mech- scenarios. They concluded that detection and containment anisms as a defense to combat the propagation of Inter- must be initiated within minutes for such systems to be ef- net worms. Rate limiting is a mechanism by which an el- fective. Singh et al. [12] proposed a system for real-time de- ement in the network can restrict the rate of communication tection of unknown worms using traffic analysis and content with other elements. Since worms spread rapidly via fast signatures. Zou et al. [18] proposed to monitor unused ad- connections to uninfected machines, rate limiting can help dress space on ingress and egress routers to detect worms at suppress the propagation of the worm. A number of rate their early propagation stage. limiting schemes have been proposed in the literature, in- Our work differs from previous works in that we focus cluding Williamson’s virus throttle [17] and Ganger’s DNS on analysis of rate control. As we demonstrate in Sections 5 based scheme [5]. However, it is not known precisely how and 6, rate control mechanisms can be extremely effective and where rate control mechanisms should be deployed in a in curtailing worm spread if deployed correctly. network. Clearly, instrumenting rate control on every indi- The primary contribution of our work is the analysis vidual node in a network is expensive administratively and of different deployment strategies for rate control mecha- hence not feasible. The question then becomes: are there al- nisms. Williamson [17] proposed the idea of host-based rate ternative deployment strategies that can yield a more desir- limiting by restricting the number of new outgoing connec- able effect than others? tions. Ganger et al. [5] proposed a scheme that analyzes and We believe that the answer to this question is yes. In limits network traffic based on abnormal DNS lookup pat- this section we illustrate the effect of different deployment terns. Both of these schemes are host based and did not ex- strategies using a star graph topology. Consider a star graph plore other deployment options. where a central hub node is connected to all the leaf nodes. We analyze two deployment scenarios: a) rate control at a 3. Background—epidemiological models certain percentage of the leaf nodes, and b) rate control at the center hub node only. Note that a star topology is very In this section we briefly introduce one class of epidemi- different from the Internet’s topology and the study of a star ological models, namely homogeneous models. Homoge- topology is mainly for demonstration of the difference from neous models are widely used in the studies of human infec- deployment at leaf and hub nodes.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    10 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us