CS – 558 paper report Name : George Christou Scalability , Fidelity and Containment in the Potemkin Virtual Honeyfarm Internet malware is a pressing concern due to the rapid evolution of large scale worms , viruses and botnets. Large scale host exploitation is responsible for : 1) *SPAM 2) *Denial of service 3) *Phishing 4) *Piracy 5) *Identity theft In addition commodity software exploits are used in order to create botnets that are used leased and sold for illegal purposes Tactical Counter measures are like spam filters and DoS defences can be employed but in order to combat the underlying infestation requires the understanding of the means and methods used to compromise and control the botnet. The principal tools for that tasks are the honeypots , which are network connected systems that are unprotected and carefully monitored in order to detect and analyze intrusions and provide data that are used to generate antivirus signatures , disinfection procedures . Also they can give information that can be used for criminal prosecution. A large number of honeypots is called a honeyfarm. Honeypot systems designs oscillate between scalability and fidelity. Low interaction : Can monitor a large fraction of a subnet but they have low fidelity and don’t execute any code. High interaction : Emulate an entire system but require more resources and can’t scale well. Another challenge of honeypot design is the containment policy i.e. prevent the compromised honeypots from attacking third party systems. The purpose of Potemkin Honeyfarm architecture is Large scale and High fidelity. This is accomplished by dynamically binding physical resources only for the short periods of time necessary to emulate the execution behavior. Thus exploiting net idleness and physical memory coherence. Based on a specialized network gateway Xen virtual machine monitor At network layer , flows are dispatched to a collection of honeyfarm servers then new VMs are dynamically instatiated for each destination IPs Fast boot is achieved by flash cloning in which wee clone an existing running vm instance and delta virtualization i.e. copy a memory page only if its changed otherwise it’s the same for all instances. Large – scale honeyfarms require a containment policy. After compromisation a honeypot may attempt to attack a third party . A weak containment may accelerate a network worm. On the other hand strict policies may blind the honeyfarm e.g. botnets need to phone home. Potemkin policy: When outbound traffic cannot be safely forwarded to the internet the gateway reflects it back to the honeyfarm , which will adopt the destination . Avoidance of resource starvation wrings for careful reflection. The main component is the gateway router : Directs inbound traffic to individual honeyfarm servers Manages resources in the long – term Responsible for outbound traffic containment policy Interface for detection and analysis In order to acquire inbound traffic , the honeyfarm may act as last-hop via BGP advertisation. The main drawback of this approach is that it renders the location visible to anyone with trace route Another solution is to configure external routers to tunnel packets destined for a particular address range back to the gateway. Packets destined to inactive IPAs are sent to non overloaded servers either randomly , or with using infection probability or even better via a type map i.e. providing the illusion that the IP hosts a particular software configuration . The latter is ideal for pre-scanned attacks. Limitations and challenges : Honeypot detection Attackers may attempt to explicitly detect that they are running in a honeypot environment and evade detection. Attackers can detect virtualized honeypots Denial of service: Blanketing the honeyfarms address space with seemingly distinct attacks may cause physical resource starvation. Modifying large number of pages can violate copy on write.