Performance Comparison of DTN Bundle Protocol Implementations ottner , Johannes Morgenroth, Sebastian Schildt, Lars Wolf Wolf-Bastian P¨ September 23, 2011: CHANTS Workshop, Las Vegas, NV
Motivation Several Bundle Protocol (BP) implementations for Linux exist DTN2: DTNRG’s reference implementation IBR-DTN: Implementation for embedded OpenWRT devices ION: JPL’s implementation for spacecrafts running on RTOS Postellation: Viagenie’s closed-source (?) BP implementation Spindle: Closed-source BP implementation by Raytheon BBN Technologies Performance? No quantitative evaluation and comparison has been performed (yet) P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 2
Motivation (cont’d) Why is performance evaluation of BP implementations important? Contacts are rare and short, communication has to be efficient Broken interoperability can point out ambiguities in the specs Goals of this paper Concentrate on open-source TCPCL implementations for Linux Quantitative “test bench” performance evaluation Quantitative opportunistic performance evaluation Systematic evaluation of interoperability P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 3
Metrics Application Layer Throughput For different storage backends Between different implementations In an opportunistic scenario Interoperability Can implementations discover each other? Throughput between different implementations Communication Efficiency In an opportunistic scenario P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 4
Experimental Setup Hardware Athlon II X4 2 . 8 GHz CPUs with 4 GiB RAM 1 GBit Ethernet (good for 940 MBit/s TCP goodput with iperf) Different Ethernet ports for measurement and control traffic Samsung Spinpoint F3 500GB (HD502HJ) hard drive Software Ubuntu Linux 11.04 DTN2 v2.7, IBR-DTN v0.6.3, ION v2.4.0 Connection blocked via iptables (IBR-DTN) or daemon’s control interface (DTN2, ION) P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 5
Storage Backends: BP Throughput 1000 Bundle Throughput [MBit/s] Bundle Throughput [MBit/s] 800 600 400 200 0 1e+07 1 10 100 1000 10000 100000 1e+06 1e+07 Bundle Payload Size [bytes] DTN2 Disk IBR-DTN Disk DTN2 Mem IBR-DTN Mem P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 6
Interoperability: Discovery, Bundle Exchange Discovery Capabilities DTN2 uses a “proprietary” discovery beacon format IBR-DTN supports IPND v0 & v1 and DTN2 discovery format ION is focussed on scheduled contacts only Interoperability Evaluation Results DTN2 and IBR-DTN discover each other just fine Bundles can be exchanged in both directions ION exchanges bundles with DTN2 and IBR-DTN P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 7
Interoperability: BP Throughput Throughput (memory storage, 1 MByte payload size) ❍❍❍❍❍ RX DTN2 IBR-DTN ION TX ❍ 687 MBit/s 635 MBit/s 93 MBit/s DTN2 881 MBit/s 679 MBit/s 90 MBit/s IBR-DTN 872 MBit/s 926 MBit/s 449 MBit/s ION Observations DTN2 may have a bottleneck in the TX component IBR-DTN and ION show unclear bottlenecksshows unclear bottlenecks Fastest result is between ION and IBR-DTN P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 8
Opportunistic Scenario: Setup Scenario A M B Memory-based storage backends A creates bundles for B at constant rate, M commutes from A to B TCP throughput limited to 12 . 7 MBit/s (based on measurements) Cycle: 72 s contact time, 30 s travel, 72 s contact time, 30 s travel (almost) 9 complete cycles in 30 minutes measurement time P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 9
Opportunistic Scenario: Transferred Data 1000 800 Transferred Data [MB] 600 400 200 0 250000 500000 1e+06 2e+06 Bundle Payload Size [bytes] DTN2 theoretical Maximum IBR-DTN P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 10
Opportunistic Scenario: Communication Efficiency 100 Communication Efficiency [%] 80 60 40 20 0 250000 500000 1e+06 2e+06 Bundle Payload Size [bytes] DTN2 theoretical Maximum IBR-DTN P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 11
Conclusions Mail me: poettner@ibr.cs.tu-bs.de Motivation Performance of BP implementations is important! Comparison of three BP implementations: DTN2, IBR-DTN, ION Results All three implementations are interoperable Performance varies widely and unpredictably Fastest result between ION and IBR-DTN Working with the implementations is not easy More results More results in our technical report (WiP, look for updates) P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 12
BACKUP SLIDES BACKUP SLIDES P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 13
Throughput (linear y) 1000 800 Bundle Throughput [MBit/s] 600 400 200 0 1 10 100 1000 10000 100000 1e+06 1e+07 Bundle Payload Size [bytes] DTN2 Disk IBR-DTN Disk ION Disk (T) DTN2 Mem IBR-DTN Mem ION Mem (T) P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 14
Throughput (logarithmic y) 1000 100 Bundle Throughput [MBit/s] 10 1 0.1 0.01 0.001 1 10 100 1000 10000 100000 1e+06 1e+07 Bundle Payload Size [bytes] DTN2 Disk IBR-DTN Disk ION Disk (T) DTN2 Mem IBR-DTN Mem ION Mem (T) P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 15
DTN2 Slowdown 250 200 Throughput [MBit/s] 150 100 50 0 1 2 3 4 5 6 7 8 9 10 Measurement Run 50e+03 Bytes Payload 500e+03 Bytes Payload 100e+03 Bytes Payload 1000e+03 Bytes Payload P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 16
Interoperability: BP Throughput Throughput (memory storage, 100 KByte payload size) ❍❍❍❍❍ RX DTN2 IBR-DTN ION TX ❍ 197 MBit/s 193 MBit/s 72 MBit/s DTN2 677 MBit/s 542 MBit/s 76 MBit/s IBR-DTN 455 MBit/s 420 MBit/s 268 MBit/s ION Throughput (memory storage, 1 MByte payload size) ❍❍❍❍❍ RX DTN2 IBR-DTN ION TX ❍ 687 MBit/s 635 MBit/s 93 MBit/s DTN2 881 MBit/s 679 MBit/s 90 MBit/s IBR-DTN 872 MBit/s 926 MBit/s 449 MBit/s ION P¨ ottner, Morgenroth, Schildt, Wolf | Bundle Protocol Performance | 17
Recommend
More recommend
