5/22/2008RMJ-1

DINETDeep Impact Network Experiment

Adapted from Technical Summary, Management and Project

Engineering

DINET CDR

6/2/08Ross Jones

5/22/2008RMJ-2

DINET Summary

• DINET is a technology validation experiment of JPL’s implementation of Delay-Tolerant Networking protocols.

• The DINET development is to produce a version of JPL’s implementation of Delay-Tolerant Networking protocols in flight and ground SW at TRL 8.

– The DINET SW is to be of sufficient quality that future flight projects can easily use it at low risk.

• DINET is to be implemented on the Deep Impact spacecraft and is being closely coordinated with the EPOXI project.

• DINET operations will be performed during the Deep Impact spacecraft team “stand down” after EPOCH operations and before the start of development for DIXI operations, i.e. Oct, 2008

– DINET requires no trajectory change. DINET developments and operations will be on a non-interference basis with EPOXI to the maximum extent possible.

Note: DINET data traffic will be AMS messages containing small images.

5/22/2008RMJ-3

Earth

Mars Phobos

OrbiterRelay

(surface asset) (surface asset)

Basic Experiment Network Topology

(DI s/c acts as orbiter relay)

5/22/2008RMJ-4

Load/Go

Deep Impact DSOT DINET EOC in PTL

7

3

2

5

4

16

6

10

stotEVRs

“Earth”

“Mars”

“Phobos”

Experimentdatabase

Load/GoLoad/Go

bundles

log msgsBRS

TCP

space links

image files

image files

LTP/UDP

8

12

20

serverclient

serverclient

serverclient

Experiment 1: Send images from nodes 12 to node 8 via nodes 6, 3, 7 (the Deep Impact spacecraft), 2, 4. Also send images from nodes 20 to node 8 via nodes 10, 5, 7 (the Deep Impact spacecraft), 2, 4.

NOTE: Deep Impactscience spacecraftis functioning as arouter (infrastructure).

5/22/2008RMJ-5

Load/Go

Deep Impact DSOT DINET EOC in PTL

7

3

2

5

4

16

6

10

stotEVRs

“Earth”

“Mars”

“Phobos”

Experimentdatabase

Load/GoLoad/Go

bundles

log msgsBRS

TCP

space links

image files

image files

LTP/UDP

8

12

20

serverclient

serverclient

serverclient

Experiment 2: Send Load/Go directive loads from node 16 to node 12 via nodes 4, 2, 7, 3, 6. Also from 16 to 20 via 4, 2, 7, 5, 10.

5/22/2008RMJ-6

Load/Go

Deep Impact DSOT DINET EOC in PTL

7

3

2

5

4

16

6

10

stotEVRs

“Earth”

“Mars”

“Phobos”

Experimentdatabase

Load/GoLoad/Go

bundles

log msgsBRS

TCP

space links

image files

image files

LTP/UDP

8

12

20

serverclient

serverclient

serverclient

Experiment 3: Omit a contact between 7 and 5 and repeat, forcing images from 20 to travel via 10, 6, 3, 7, 2, 4 and forcing directive loads to 20 to travel via 4, 2, 7, 3, 6, 10.

XNOTE: spacecraft is temporarily unable to function as a router.

5/22/2008RMJ-7

Load/Go

Deep Impact DSOT DINET EOC in PTL

7

3

2

5

4

16

6

10

stotEVRs

“Earth”

“Mars”

“Phobos”

Experimentdatabase

Load/GoLoad/Go

bundles

log msgsBRS

TCP

space links

image files

image files

LTP/UDP

8

12

20

serverclient

serverclient

serverclient

Experiment 4: manually route traffic between nodes 12 and 20 (both remote) via the orbiter, without Earth in the loop from node 20 to 12 and 12 to 20.

5/22/2008RMJ-8

The DINET Stack

UT adapter

TM/TC

R/F, optical

LTP

BP DTN forwarding

spacepackets

AMSmessaging

Remote AMScompression

Convergence layer adapter

CFDP File Data PDUs

(“Protocol X”)

Image publisher/receiver

load/go

rfx,admin

programs,clocks

5/22/2008RMJ-9

Interplanetary Overlay Network (ION)

• Reference implementation for the DTN Bundle Protocol (BP) is DTN2, maintained at UC Berkeley.– Designed as a research vehicle.– Widely used, well supported.

• Most DTN researchers are investigating terrestrial applications, for which DTN2 works very well.

• Space flight applications impose different constraints, motivating development of an alternative BP implementation for use in space flight missions.

• ION is an implementation of BP/LTP, developed at JPL, that’s designed to be usable in flight.

5/22/2008RMJ-10

Constraints on a Flight Implementation

• Link constraints– All communications are wireless, generally slow,

asymmetric.• From spacecraft to ground: 256 Kbps to 6 Mbps.• From ground to spacecraft: 1 to 2 Kbps.

– Links are very expensive, virtually always oversubscribed.– Fine-grained data delivery.

• Immediate delivery of partial data is often OK.

• Processor constraints– Flight processors typically run real-time operating systems

(VxWorks®, RTEMS™) lacking protected memory models.– Robustness is paramount. No malloc and free or standard

new and delete; must not crash other flight software.– Processing efficiency is important:

• Slow (radiation-hardened) processors.• Relatively slow non-volatile storage: flash memory.

5/22/2008RMJ-11

ION’s Divergence From DTN2Design

ElementDTN2 ION Rationale

Language C++ C Processing efficiency, memory management visibility.

Memory management new, delete PSM No dynamic system memory management permitted.

Non-volatile storage management

Berkeley DB, RDBMS (MySQL)

SDR persistent objects

Processing efficiency, footprint.

Locus of processing dtnd daemon process, separate routing engine

highly distributed: forwarders, ducts, applications, and admin tools

Robustness (module simplicity, incremental upgrade; prevent head-of-line blocking); simplify flow control.

Locus of node state (e.g., queues)

private memory of dtnd daemon

shared memory Support distributed functionality, limit impact of demand spikes.

Application Programming Interface

remote procedure calls to dtnd

shared library functions act on shared memory

Support real-time operations: prevent blocking, support deterministic execution.

Endpoint IDs in bundle’s primary block

only ASCII URIs in dictionary

supports CBHE Bandwidth efficiency.

5/22/2008RMJ-12

Performance

DTN Bandwidth Tests

1 10 100

500750

1024

5120

102401536020480

51200 10240020480025600

0

100

200

300

400

500

600

700

800

900

1.00E+00 1.00E+01 1.00E+02 1.00E+03 1.00E+04 1.00E+05 1.00E+06Data Size (Kilobytes)

Ave

rag

e D

ata

Rat

e (M

bit

s/se

con

d)

2 nodes DTN2 RAM

3 nodes DTN2 RAM

2 nodes DTN2 disk

3 nodes DTN2 disk

2 nodes ION RAM

3 nodes ION RAM

ION flight software footprint: about 708 kilobytes including SDR database management system.

5/22/2008RMJ-13

Contact Graph Routing#*****************************************************************************## ** DINET experiment pass #1. Monday morning, October 20. **#@ 2008/10/20-11:00:00a range +0 +14400 2 7 79a range +0 +14400 3 7 79a range +0 +14400 5 7 79## Contact between nodes 5 and 7 for 60 min.@ +0a contact +0 +3600 5 7 250a contact +0 +3600 7 5 20000## Contact between nodes 3 and 7 for 120 min.@ +3900a contact +0 +7200 3 7 250a contact +0 +7200 7 3 20000## Contact between nodes 2 and 7 for 50 min.@ +7500a contact +0 +3000 2 7 250a contact +0 +3000 7 2 20000#

5/22/2008RMJ-14

Status of ION

• Conforms to version 6 of the BP specification (June 2007).

• Single code base runs without modification in all environments. So far:– Red Hat Linux 8+, Ubuntu Linux on 32-bit processors.– Fedora Core 3+, on 32-bit and 64-bit processors.– VxWorks 5.4 on PowerPC 750.– Mac OS/X

• Interoperability with DTN2 (and other Bundle Protocol implementations: C#, .Net, Symbian) demonstrated at IETF in San Diego, November 2006.

5/22/2008RMJ-15

DINET Instrumentation

• Protocol status, diagnostic, and statistics messages issued by every node, including the spacecraft.

• Current network topology and running logs of messages displayed on the operations console in the Experiment Operations Center.

• Detailed “watch” character stream of event indications can be selectively enabled and disabled in real time at each node.

5/22/2008RMJ-16

Key Metrics

• Metric 1 – Link Utilization Rate• Metric 2 – Delivery Acceleration ratio• Metric 3 – ION Node Storage Utilization • Metric 4 – Multipath Advantage

• Priority X • Dynamic Routing X• Automated Forwarding X• Custody Transfer X X• Delay-Tolerant Retransmission X • Flow & Congestion Control X X

Link Utilization

DeliveryAcceleration

Ratio

ION Node Storage Utilization

MultipathAdvantage

Applicability to DTN Features

5/22/2008RMJ-17

DTN Validation Criteria• Metric 1 – Path utilization rate (U)

– U = RT/K, where RT is total volume of science data returned and K is the total data return capacity (adjusted per artificially induced segment loss as applicable).

– Measures the effectiveness of automatic forwarding, custody transfer, and delay-tolerant retransmission.

– Validation criteria:• Ua > 90%. (DTN uses the links efficiently when there is no induced data loss.)

• Ub > 90%. (DTN remains efficient despite an increase in the rate of data loss .)

• Metric 2 – Delivery acceleration ratio (G)– W = (.5 *) + (1.0 * R1) + (2.0 * R2), where W is the urgency-weighted volume of science

data returned and R0, R1, and R2 are respectively the total volumes of priority-zero, priority-1, and priority-2 science data returned. (Note that RT = R0 + R1 + R2.)

– Q0 = .25 * RT where Q0 is the “reference” volume of priority-zero science data returned, so computed because we will arbitrarily assign priority zero to 25% of all DINET science data.

• Similarly, Q1 = .60 * RT and Q2 = .15 * RT. – V = (.5 * Q0) + (1.0 * Q1) + (2.0 * Q2), where W is the urgency-weighted reference

volume of science data returned.– G = W / V.– Measures the effectiveness of the priority system.– Validation criteria:

• Ga > 1. (Prioritization accelerates the delivery of urgent data.)• Gb > 1.

5/22/2008RMJ-18

DTN Validation Criteria (Continued)• Metric 3 – ION node storage utilization

– Retention of a stable margin of unassigned space at each node measures the effectiveness of congestion control.

– Validation criteria:• The total number of bundles for which custody is refused anywhere in the

network for the reason Depleted Storage, throughout each experiment, is always zero. (Never run out of storage anywhere.)

• NX4 = NX3 and NX8 = NX7 for all values of X. (Storage utilization stabilizes over the course of the experiment.)

• Metric 4 – Multipath advantage– The net path capacity PXYa for any single path from node X to node Y during

configuration a is the smallest value of ∑KABZ for Z = 14 among all links (A, B) in that path; PXYb is similarly defined for configuration b.

– The multipath advantage MXYa for traffic from X to Y during configuration a is computed as ∑PXYa for all paths from X to Y, divided by the largest single PXYa among all paths from X to Y, minus 1.

– Where there is only a single possible path between X and Y, multipath advantage is zero. Multipath advantage measures the effectiveness of dynamic routing.

– Validation criteria:• MXYa > 0 for X = node 20 and Y = node 8. (Dynamic routing among multiple possible

paths increases the total network capacity from Phobos to Earth.)• MXYb > 0 for X = node 20 and Y = node 8..

5/22/2008RMJ-19

Environment Envelope• Given or measured quantities that will be reported as part of the experiment• These are the primary mission parameters and initial conditions that affect the

performance results for a given DTN implementation.

• Environment Envelope– Propagation Delay - 2min– Partition Delay - 5 day (contact latency - network is partitioned)– File Size - Range is 2-65KB

• Maximum size message with AMS is 65KB• For images larger than 65KB, a mission should use CFDP in conjunction with AMS and BP but this

is not part of our experiment since simple unacknowledged CFDP has been implemented previously.

– Data Rates (128-400,000 bps)– Number of end nodes (11)– Number of links per node– total number of links– Contact duration (4 hours)– Data volume– Data Completeness– Data Quality– Bit Error Rate– Available buffer size

5/22/2008RMJ-20

DTN Protocol Envelope

5/22/2008RMJ-21

Project schedule

• Critical path runs through the ION/DIAS testing on EPOXI test beds

• There are 17 days of funded project schedule reserve