Simulation of DNS(Domain Name System) Using SimLib

Transcription

1 Simulation of DNS(Domain Name System) Using SimLib Submitted by Prem Tamang Submitted to Dr. Lawrence J. Osborne

2 Table of Contents 1. Introduction 3 2. Motivation and Challenges Assumptions 5 4. Implementations.. 7 Event Graph Selection of Input Probability Distribution for the Random variables existing in the Model i) Assessment of Sample Independence ii) Hypothesizing Families of Distributions iii) Estimation of Parameters iv) Determining How Representative the Fitted Distribution are v) Goodness-of Fits Tests 5. Simulation Results Analysis of results References.. 18

3 1. Introduction The Domain Name System (DNS) is a distributed database that allows local control of the segments of the overall database, yet data in each segment are available across the entire network through a client-server scheme. This hierarchical naming system contains programs called name servers which stores information about some segment of the database and make it available to clients, called resolvers. Resolvers are often just library routines that create queries and send them across a network to a name server. This project simulates the behavior of the name servers which resolves domain names meaningful to humans into the numerical (binary) identifiers associated with networking equipment for the purpose of locating and addressing these devices worldwide. Client1 Client2 Client3 Client4 Client5 NS Queuries[e.g.google.com] QUEUE NS Queuries[e.g.google.com] NameServer1 NameServer 2 Reply[COM:IP Address] Root General Top Level Domaim COM Name Servers EDU Microsoft.com Novell.com Google.com Query Resolved [e.g.google.com : ] Model of the Simulation System Lamar.edu Tamu.edu Gatech.edu This simulation system consists of the name servers that receive requests one at a time from the set of clients within its network in a continuous manner. Name servers are present in three levels, the root level, the top level domains and the respective authoritive name servers. Name Servers

4 supports Iterative Queries or Non-Recursive Queries i.e. the requester must be capable of pursuing referrals. The requests from the clients are queued in the name server if the name server is busy in processing the packet. Once the request packet is processed, the query will passed to the Root level Domain which contains information regarding all the top level domains like com, edu, info etc. Then the root level domain refers the respective top level domain (i.e. provides the IP Address) to the names server.the request will again be queued in the name server since the packets needs to be processed. The request will be redirected to the top level domains which again sends the referrals(ip Address) of the respective Name Server. Finally the respective Authoritive name servers provides the IP Address for the given domain name back and the response will be forwarded back to the Client. e.g. Client 1 sends the request to the NameServer1. There exist equal chances of selection of the NameServers based on the equal probability. If NS1 is busy processing the packet, it will be stored into the queue. Then the reqest is forwarded to the Root Level Domain [XX.XX.XX.XX]. It replies back with the IP Address of the Top Level Domains - COM. The request are again queued on the NS1 server. Once its turns comes, again the request will be forwarded for resolving to the COM Top Level Domain. The COM Top level domains again replies back with the IPAddress of the NameServer of Google.com. The request is again processed at NS1 and the redirected to the NameServer of the Google. Finally the Authoritive NameServer of google.com provides the IP Address of the WebServer in the DNS Reply to the NS1. The NS1 then replies back to the client with the valid IP Address of the requested domain

5 2. Motivation and Challenges In the internet, DNS is one of the mostly used systems since for accessing any resource on the internet, it requires the IP Address of that resource and DNS provides this. Every network consists of at least one DNS Server. The performance of the network is heavily dependent upon the Name Servers. So it s always important to know how well the DNS performs in the network. Various factors like the total number of queries, type of queries, the response time, and numbers of available master DNS Servers affects the performance of DNS Servers. So this project provides the simulation model of the DNS system to measure the arrival rate of the DNS queries, response time of the Queries, handling of various types of valid and invalid queries and load balancing depending upon the numbers of DNS Servers etc. The biggest challenge in this project was hypothesizing the probability distribution for the random variable like Inter-arrival Time of DNS Queries, Service type for the packet processing, Round Trip time to the Domain Name Servers. Also, designing the work-flow model for representing the name servers and various zones was quite difficult. 3. Assumptions: The system consist of basic components only i.e. Client Machines, Name Servers. The zone contains three levels, Root zone, general Top Level Domain(gTLD) and the authoritive name servers. Currently the system implements only two types of gtld, COM and EDU. Each gtld at least contains the name server information of 5 authoritive name servers. NameServer supports only Iterative Queries and doesn t provide the caching. Interarrival time of the DNS Queries are IID and arrives the system one at a time. Data for estimating the Mean Value for Interarrival time has been obtained from the internet sites [2][3].. The servers takes time ( though very less) for the packet processing and if the server is busy processing the packet, the request from the client as well as the responses from the other root, gtld and authoritive name servers has to wait in queue.

6 Response time i.e. the total time when the client gets back the response for this DNS Queries is dependent upon the Service Time or Packet Processing Time of the DNS Server, Round Trip Time from the DNS Server to the zone servers. The simulation is performed for length of 5 hrs. List Specifications: The simulation model measure the performance of the DNS on the basis of following measures : Average Response Time of the System Average Delay in the Queue Time Average No. of the Request and Responses in the Queue Utilization of the Name Servers Efficiency of the system depending upon the various numbers of name servers Proportion of the DNS Queries for Invalid Domains Proportion of the NDS Queries for COM and EDU Domains.

7 4. Implementation : This project uses the C programming language and SIMLIB package for simulation. Arrival Departure End Simulation Event Graph for DNS System Model. Selection of Input Probability Distribution : This simulation model basically consists of three Random Variables. i) Inter-arrival Time ii) Service Time iii)round Trip Time All the random variables are Identical- Independent Distribution. This project uses the standard techniques of statistical inference to fit a theoretical distribution form to the data and to perform hypothesis tests to determine the goodness of fit. The following steps are performed on the collected data by the help of Expertfit Software from Averill M.Law & Associates to select the Input Probability Distribution:

8 Data Summary from Expertfit Software. i) Inter-arrival Time : Hypothesized Probability Distribution : Weibull Probability Distribution a) Assessment of Sample Independence Scatter Diagram : Lag-Correlation Plot:

9 b) Hypothesizing Families of Distributions: Histograms: c) Estimation of Parameters : The probability distribution hypothesized appears to be Gamma distribution which belongs to the Heavy Tailed Distribution Family. Estimated Parameters :

10 Alpha = Shape Parameter = Beta = Scale Parameter = d) Determining How Representative the Fitted Distribution are : Density-Histogram Plots : Distribution-Function-Differences Plots : e) Goodness-of Fits Tests : H 0 : The X i s are IID random variables with Weibull Probability Distribution function

11 Chi-Square Tests : Kolmogorov-Smirnov Tests : Thus the null Hypothesis H 0 cannot be rejected. The random variates in this simulation model are generated from the probability distribution, Weibull, using the following formula. X=B(-1n U) 1/α where U ~ U(0,1)

12 ii) Service Time or Packet Processing Time : Actual sets of data could not be achieved. However based on the interpreted data on site [4] the value considered for service time is microseconds. So a uniform distribution U( ) is used to generate the Random Variates. iii) Round Trip Time : Actual sets of data could not be achieved. However, the gtld Server [5] shows that the RTT time varies about ms. So a uniform distribution is considered U( ) to generate the Radom Variates. 5. Simulation Results: The simulation is performed with the varying numbers of clients and servers. When the numbers of Servers = 2 and No. of Clients machines is varied from 5 to 15 DNS Server Simulation ===================== Total Number of Server : 2 Number of Client Machines : 5 to 15 by 10 Shape Parameter(Alpha) for Weibull Distribution of Interarrival Time : seconds Scale Parameter(Beta) for Weibull Distribution of Interarrival Time : seconds Lower Limit for Uniform Distribution of Service Time : Upper Limit for Uniform Distribution of Service Time : Lower Limit for Uniform Distribution of Round Trip Time : Upper Limit for Uniform Distribution of Round Trip Time : Total Time for Simulation : minutes seconds seconds seconds seconds No. of Clients : 5 Proportion of DNS Queries Related to COM Domain : (183/564) : Proportion of DNS Queries Related to EDU Domain : (192/564) : Proportion of Invalid Queries : (189/564) : Response Times for Client Machines ===================================

13 sampst Number variable of number Average values Maximum Minimum CPU Utilization ================= File Time number average Maximum Minimum Time Average Number of Delay for Requests and Responses in Queue ================= File Time number average Maximum Minimum E E No. of Clients : 15 Proportion of DNS Queries Related to COM Domain : (557/1676) : Proportion of DNS Queries Related to EDU Domain : (565/1676) : Proportion of Invalid Queries : (554/1676) : Response Times for Client Machines =================================== sampst Number variable of number Average values Maximum Minimum

14 CPU Utilization ================= File Time number average Maximum Minimum Time Average Number of Delay for Requests and Responses in Queue ================= File Time number average Maximum Minimum When the numbers of Servers = 5 and No. of Clients machines is varied from 5 to 15 DNS Server Simulation ===================== Total Number of Server : 5 Number of Client Machines : 5 to 15 by 10 Shape Parameter(Alpha) for Weibull Distribution of Interarrival Time : seconds Scale Parameter(Beta) for Weibull Distribution of Interarrival Time : seconds Lower Limit for Uniform Distribution of Service Time : seconds Upper Limit for Uniform Distribution of Service Time : seconds Lower Limit for Uniform Distribution of Round Trip Time : seconds Upper Limit for Uniform Distribution of Round Trip Time : seconds Total Time for Simulation : minutes No. of Clients : 5 Proportion of DNS Queries Related to COM Domain : (183/564) : Proportion of DNS Queries Related to EDU Domain : (192/564) : Proportion of Invalid Queries : (189/564) : Response Times for Client Machines =================================== sampst Number variable of

15 number Average values Maximum Minimum CPU Utilization ================= File Time number average Maximum Minimum Time Average Number of Delay for Requests and Responses in Queue ====================================================== File Time number average Maximum Minimum E E E E E E E E E+30 No. of Clients : 5 Proportion of DNS Queries Related to COM Domain : (557/1676) : Proportion of DNS Queries Related to EDU Domain : (565/1676) : Proportion of Invalid Queries : (554/1676) : Response Times for Client Machines ============================ sampst Number variable of number Average values Maximum Minimum

16 CPU Utilization ============ File Time number average Maximum Minimum Time Average Number of Delay for Requests and Responses in Queue ====================================================== File Time number average Maximum Minimum E E E E E

17 6. Analysis of Results The results from the simulation model are obtained as expected. As we can see that, keeping the number of servers constant (i.e. 2) in first case, when the no. of clients are increased from 5 to 15, CPU Utilization as well as the time average no. delay of requests and responses in the Queue are also increased. As we can see the packet processing time by the Name Server in comparison to the Round Trip Time, is very small, so we don t have many packets in the queue which also explains the low utilization value of CPU. Server simply processes the packets and redirects it to the other Domain Servers. Similarly, when the numbers of name servers are increased the load is divided resulting into low CPU Utilization as well as low value of time average no. of delay of requests and responses in the Queue. Due to the lack of the output data from the real system, the output of this simulation could not be validated to the full extend. However, regarding the Interarrival time, 100 samples of data were taken from simulation and fed into the Expertfit Software, which confirms the Weibull Distribution of the Interarrival time. Further the Confidence-Interval Approach Based could have been applied for statistical comparison of the real system observed data and simulation output data.

18 References: 1) Jaeyeon Jung, Emil Sit, Hari Balakrishnan, Reobert Morris,DNS performance and the effectiveness of caching. IEEE/ACM Transactions on Networking (TON ), pages, , ) 3) 4) 5)