Performance Analysis and Comparison of Different DNS64 Implementations for Linux, OpenBSD and FreeBSD

Transcription

1 Performance Analysis and Comparison of Different DNS64 Implementations for Linux, OpenBSD and FreeBSD Gábor Lencse and Sándor Répás Széchenyi István University Győr, Hungary

2 IPv4 Address Exhaustion The new clients will not be able to get IPv4 addresses The majority of the Internet sites still uses IPv4 protocol (and will do so for a long time) Different IPv6 transition solutions exist for different problems DNS64+NAT64 are for IPv6 clients and IPv4 servers 2

3 Related Work Other researchers: Analyze the common performance of a given DNS64 implementation + a given NAT64 implementation Our view: DNS64 and NAT64 are two independent services Their performance should be analyzed separately This paper: DNS64 (NAT64: in a separate paper) 3

4 Operation of DNS64 DNS64 is an extension of the DNS service If the symbolic name has an AAAA record then it is returned If it has an A record only then an IPv4 Embedded IPv6 Address containing the IPv4 address in its last 32 bits is generated and returned 4

5 DNS64 Implementations BIND 9.8 (Berkeley Internet Name Daemon) As a recursor As a forwarder TOTD (Trick or Treat Demon) As a forwarder Operating Systems Linux OpenBSD FreeBSD 5

6 DNS64 Test Network Tested software DNS64 implementations running on a low performance computer (Pentium III, 800MHz) Other devices 8 Clients (providing requests) Authoritative DNS Server 8+1 high performance computers (Dell Precision 490 Workstations) 3Com Baseline Switch (2948-SFP Plus) Authoritative DNS server teacherb.tilb.sze.hu 2001:738:2c01:8001::111/64 client1 DNS64 server Dell Precision 490 Intel PIII 800MHz... client computers for all the tests 8x Dell Precision / / / :738:2c01:8001::1/64 3com Baseline Switch 2948-SFP Plus 2001:738:2c01:8001::118/64 client8 6

7 DNS64 tests done by using the host Unix command resolving systematically generated symbolic names: 10-{0..10}-{0..255}-{0..255}.zonat.tilb.sze.hu mapped to the IPv4 addresses by DNS and to the 2001:738:2c01:8001:ffff:ffff:0a00: :738:2c01:8001:ffff:ffff:0a0a:ffff IPv6 addresses by the DNS64 implementations 7

8 Script for an experiment An experiment contained the execution of 256 name resolutions using the host command file: dns-st-c.sh #!/bin/bash for c in { } # that is 64 iterations do host 10-$1-$2-$c.zonat.tilb.sze.hu & host 10-$1-$2-$((c+1)).zonat.tilb.sze.hu & host 10-$1-$2-$((c+2)).zonat.tilb.sze.hu & host 10-$1-$2-$((c+3)).zonat.tilb.sze.hu done 8

9 Script for a client Each client performed 256 experiments #!/bin/bash i=`cat /etc/hostname grep -o.$` # i: {1..7} rm dns64-$i.txt do for b in {0..255} do /usr/bin/time -f "%E" -o dns64-$i.txt \ a./dns-st-c.sh $i $b done done 9

10 Measurements To simulate different measures of load, the number of clients were: 1, 2, 4 and 8 Measured values: Execution time of a single experiment (256 host commands) CPU utilisation of the test computer Memory consumption of the given DNS64 implementation (measured by the change of the amount of the free memory not exact!!!) 10

11 DNS64 Performance Results: BIND, Linux, Recursor 1 Number of clients Average exec. time of 256 host commands (s) standard deviation maximum value CPU utilization (%) standard deviation DNS64 memory consumption (MB) Number of requests served (request/s)

12 DNS64 Performance Results: BIND, OpenBSD, Recursor 1 Number of clients Average exec. time of 256 host commands (s) standard deviation maximum value CPU utilization (%) standard deviation DNS64 memory consumption (MB) Number of requests served (request/s)

13 DNS64 Performance Results: BIND, FreeBSD, Recursor 1 Number of clients Average exec. time of 256 host commands (s) standard deviation maximum value CPU utilization (%) standard deviation DNS64 memory consumption (MB) Number of requests served (request/s)

14 Comparison of the exec. time of 256 host commands, BIND, Recursor 14

15 DNS64 Performance Results: TOTD, Linux, Forwarder 1 Number of clients Average exec. time of 256 host commands (s) standard deviation maximum value CPU utilization (%) standard deviation DNS64 memory consumption (MB) Number of requests served (request/s)

16 Conclusions/Recommendations 1 BIND9 running under Linux showed the best overall performance characteristics from among the tested DNS64 solutions. It was stable under serious overload conditions and its memory requirement was found moderate. BIND9 running under Linux is our number one recommendation for a DNS64 solution in a production system. 16

17 Conclusions/Recommendations 2 BIND9 running under FreeBSD was very stable but showed less good performance than under Linux, due to running in jail. It can also be a choice if FreeBSD platform is preferred for security reasons. 17

18 Conclusions/Recommendations 3 The memory usage of TOTD running under all the three operating systems was very low thus it can be a good DNS64 solution for small embedded systems but it needs an external tool for load limitation otherwise it may lose its stability. Due to its average performance, TOTD would deserve a thorough code review and bug fix. It is also a good candidate for a client side DNS64 solution. 18

19 Thank you for your attention! Questions? 19