QoS Evaluation Model for a Campus-Wide Network: an alternative approach Juan Antonio Martínez Comunicacions - Servei d’Informàtica Universitat Autònoma de Barcelona

Index Objectives Objectives Classical approach to QoS Classical approach to QoS Evaluating network availability: a generic model Evaluating network availability: a generic model Extending the model to the services Extending the model to the services Practical results Practical results Comparison to other existing products and future work Comparison to other existing products and future work Conclusions Conclusions

Objectives Determine if our network reaches the ‘5 9s’ objective (99.999% of availability) Determine if our network reaches the ‘5 9s’ objective (99.999% of availability) Obtain a generic model to evaluate the quality of the network regarding Obtain a generic model to evaluate the quality of the network regarding  Infrastructure  Services The model must The model must  Be simple  Adapt easily to network topology changes

Obtaining the model

The classical approach Quality is evaluated as a combination of: Quality is evaluated as a combination of:  Delay  Jitter  Packet loss This approach is useful for WAN links and environments This approach is useful for WAN links and environments  Easy to measure in both router endpoints In our opinion, it is not suitable for LAN environments In our opinion, it is not suitable for LAN environments  Parameters are difficult to determine (distributed environment)  Gathered data is not significant (burst traffic)

Problems of this model Quality is not assigned a numeric value Quality is not assigned a numeric value In a LAN environment In a LAN environment  Network probes must be distributed through the network  In switched environments, each segment provides a different value  Expensive to implement  Fluctuant values due to the traffic nature In fact, there is not a model. It is just an evaluation of the aforesaid parameters In fact, there is not a model. It is just an evaluation of the aforesaid parameters

Our model:the basis Whichever the system is, it must: Whichever the system is, it must:  Be of low cost.  Provide QoS as a numerical value.  Be flexible enough to adapt itself to network topology changes. Our idea is to get data from : Our idea is to get data from :  Simple tools like ‘ping’  SNMP queries We discard complex solutions (modified TCP stacks, proprietary PING...) We discard complex solutions (modified TCP stacks, proprietary PING...)

Conceptual development of the model Assume that all network devices are known and with SNMP management facilities. Assume that all network devices are known and with SNMP management facilities. Assume that the number of users affected by any network device failure can be determined. Assume that the number of users affected by any network device failure can be determined. The idea beyond the process is The idea beyond the process is  Choose the critical network devices  Weight them accordingly  Determine the instant availability  Compute the mean value

Choosing the devices and the weights for the model

Mathematical analysis (I) Let Let  Kd be the relative availability coefficient  di be a binary value that tells whether a segment is accessible or not For the mean value this implies For the mean value this implies

Mathematical analysis (and II) If we sample at constant intervals If we sample at constant intervals For efficient computing, this means For efficient computing, this means In this way, we can evaluate the availability with the number of samples, the previous mean value and the last sample. In this way, we can evaluate the availability with the number of samples, the previous mean value and the last sample.

Extending the model to the services Our first goal is to determine whether the service is working or not (ok/not ok) Our first goal is to determine whether the service is working or not (ok/not ok) The explained network model is suitable for the services with few changes : The explained network model is suitable for the services with few changes :  Omission of the criticity values (Kc=1)  A tool to determine instant availability is needed (ping is no longer valid)

Details regarding practical implementation

Network availability Availability is computed based on ICMP tests (ping) Availability is computed based on ICMP tests (ping) Second-level granularity Second-level granularity To compute the coefficients To compute the coefficients  Kc1 : from the network topology.  Kc2 : from our experience. The global coefficient is computed as the arithmetic mean value The global coefficient is computed as the arithmetic mean value A config file stores the network devices that will be tested A config file stores the network devices that will be tested

config file example [ NETWORK] #ip_name ip_address availability coef. gw CB si0swfo1+si0swfo c7p1sw anccswit m0swit cvcsw cibibfo hottwp1+hottwp dretswit ecllefo bhum1sw ftisw recswit vetswit

Service availability The model is essentially the same, but now for each machine we analyse its critical services The model is essentially the same, but now for each machine we analyse its critical services The system can The system can  Evaluate proper function of a service at a given time  Compute the availability over time A proprietary MIB is used to determine the critical parameters of each service (SNMP queries are supported) A proprietary MIB is used to determine the critical parameters of each service (SNMP queries are supported)

Monitoring requirements A set of ‘C’ programs (one for each service) A set of ‘C’ programs (one for each service) A global configuration file: A global configuration file: [SERVICES] #host name services to be tested cc.uab.es smtp pop imap news.uab.es nntp ftp.uab.es ftp... Optionally, for the services : A proprietary MIB with the data we want to monitor (not mandatory) A modified version of the snmpd daemon

Practical Results in our Campus Network

Environment Development : Development :  PC Pentium II Linux RedHat 6.0  gcc Production: Production:  Ultra Enterprise Solaris 2.6 (gcc )  PC Pentium II Linux RedHat 6.0 (gcc ) Our network availability has achieved monthly values between and 1.0 Our network availability has achieved monthly values between and 1.0 For the services, we use it both for availability values and to test that they are working properly. For the services, we use it both for availability values and to test that they are working properly.

Overview of the system A set of routines that verify that the services are working properly: A set of routines that verify that the services are working properly:  bootp,dhcp,dns,ftp,http,smtp,pop,imap,radius,nntp A C program that implements the mathematical model A C program that implements the mathematical model Configuration file, which includes Configuration file, which includes  Coefficients for the model  Servers and services to be monitored. Optionally, a modified version of the snmpd daemon if access to the proprietary MIB is desired. Optionally, a modified version of the snmpd daemon if access to the proprietary MIB is desired.

Measured availability data

‘on line’ service monitoring Based on the programs that are used to check service availability Based on the programs that are used to check service availability Useful to network operators to detect network problems Useful to network operators to detect network problems Operating modes : Operating modes :  Interactive: provides a report with configurable debug  Cron-based : generates mail/sms messages if any problems are detected

Interactive execution example *** 14:35: blues.uab.es:smtp blues Sendmail 5.65v4.0 ( /17Dec AM) Tue, 7 Nov :44:20 + QUIT 221 blues closing connection *** 14:35: blues.uab.es:smtp 116b,0.0432s,2.6Kb/s OK! *** 14:35: blues.uab.es:pop3... +OK POP3 blues v7.63 server ready QUIT +OK Sayonara *** 14:35: blues.uab.es:pop3 49b,0.152s,0.31Kb/s OK! *** 14:35: blues.uab.es:imap... * OK blues IMAP4rev1 v server ready A01 LOGOUT * BYE blues IMAP4rev1 server terminating connection A01 OK LOGOUT completed *** 14:35: blues.uab.es:imap 121b,0.102s,1.2Kb/s OK! *** 14:35: news.uab.es:nntp news.cesca.es InterNetNews NNRP server INN ready (posting ok). QUIT 205. *** 14:35: news.uab.es:nntp 81b,0.253s,0.31Kb/s OK! *** 14:35: ftp.uab.es:ftp koala FTP server (Version wu-2.5.0(1) Wed Oct 20 12:02:15 DST 1999) ready.

Comparison to other existing products and future work

Comparison As a monitoring tool, there are other existing products, which are really powerful As a monitoring tool, there are other existing products, which are really powerful  Big Brother  NoCoL Regarding the graphic tools, they generate no data, but simply display it. Regarding the graphic tools, they generate no data, but simply display it. None of these gives a detailed model for the network availability or monthly reports None of these gives a detailed model for the network availability or monthly reports

Future work Analysis and integration of multiple meters Analysis and integration of multiple meters Improvement of the Web interface Improvement of the Web interface WAP interface to the monitoring system WAP interface to the monitoring system

Conclusions

Benefits of the system Proper working of all servers and services is easy and centrally verified Proper working of all servers and services is easy and centrally verified Easy adaptation to network changes Easy adaptation to network changes  Network devices  New servers or services Monthly reports of both internal and external connectivity Monthly reports of both internal and external connectivity Availability reports of the relevant services Availability reports of the relevant services Easy to integrate with graphic tools such as MRTG Easy to integrate with graphic tools such as MRTG

For more information... This presentation: This presentation:  ftp://ftp.uab.es/pub/terena/QoS.ppt The paper can be found at: The paper can be found at:  ftp://ftp.uab.es/pub/terena/QoS.pdf

Doubts, comments, suggestions...