A Method to Estimate the Timestamp Accuracy of Measurement Hardware - - PowerPoint PPT Presentation

a method to estimate the timestamp accuracy of
SMART_READER_LITE
LIVE PREVIEW

A Method to Estimate the Timestamp Accuracy of Measurement Hardware - - PowerPoint PPT Presentation

A Method to Estimate the Timestamp Accuracy of Measurement Hardware and Software Tools Patrik Arlos Blekinge Institute of Technology Inter-arrival time Mean: 1230 s Min: 1110 s Max: 1623 s Std.dev: 44.8 s Mean: 1230 s Min: 1230


slide-1
SLIDE 1

A Method to Estimate the Timestamp Accuracy of Measurement Hardware and Software Tools

Patrik Arlos Blekinge Institute of Technology

slide-2
SLIDE 2

2

Inter-arrival time

Mean: 1230 μs Min: 1110 μs Max: 1623 μs Std.dev: 44.8 μs Mean: 1230 μs Min: 1230 μs Max: 1721 μs Std.dev: 2.2 μs

slide-3
SLIDE 3

3

Accuracy

  • To what level can we trust a timestamp?

1117564161.171961925000 (Hardware) 1113647109.171036000000 (Software)

Many digits do not mean high accuracy

The comparison between a measured value and the correct value

slide-4
SLIDE 4

4

Timestamp

  • Resolution

T = 0.2, 0.2, 0.4, 0.5

slide-5
SLIDE 5

5

Timestamp

  • Resolution
  • Implementation

GetTimestamp() GetTimestamp()=2 trequest tresponse Clock Value 1 2 3 4

slide-6
SLIDE 6

6

Timestamp

  • Resolution
  • Implementation
  • Synchronization
slide-7
SLIDE 7

7

Timestamp Accuracy Method

  • 1. Generate traffic, fixed-size PDUs back-to-back
  • 2. Collect the traffic (PDUs) and their timestamps
  • 3. Verify that the trace is correct, i.e., no loss or

reordering has occurred

  • 4. Calculate the inter-arrival time,

5. 6.

i I

T , ˆ

C L T /

I =

I , I

ˆ T T i

i

− = ε

⎪ ⎩ ⎪ ⎨ ⎧ + + =

Δ

2 Case ) min( ) max( 1 Case 2 ) min( ) max( ε ε ε ε T

slide-8
SLIDE 8

8

Setup

slide-9
SLIDE 9

9

Results

  • Snapshots for the evaluated systems!
  • Vanilla systems, i.e. no optimization efforts were

made.

  • ntpd used the default ntp.conf of the OS/distribution.
  • ntpq was used to verify that stability was reached, before starting

the session.

  • Link: 10Mbps Full duplex Ethernet.
  • MP was a Pentium-4 2.4GHz/1GByte RAM.
  • 250 000 PDUs ~ 300s
slide-10
SLIDE 10

10

Timestamp Accuracy Estimations

410 μs 573 μs Linux 2.6 Linux 2.4 410 μs RAW-TSC 300 μs Linux 2.6 Linux 2.4 RAW-ioctl System 460 μs 340 μs 440 μs 320 μs 3.24 ms 374 μs 346 μs 100 ns 60 ns TΔ TΔ System Linux 2.6 Linux 2.4 PCAP-MMAP Linux 2.6 Linux 2.4 FreeBSD 5.1 PCAP-PFRING Linux 2.4 Linux 2.6 PCAP Agilent J6800 DAG 3.5 Pentium-4 2.4 GHz Synchronized using NTP 250000 PDUs ~300 s.

slide-11
SLIDE 11

11

Software Based systems

ε [ns]

slide-12
SLIDE 12

12

Conclusions

  • Timestamp accuracy influences ALL subsequent

analysis.

  • Synchronization is a major problem.
  • Timestamp accuracy depends on the entire

system, hence the entire system needs to be evaluated, before use.

  • HW based systems, ~1us
  • SW based systems, ~1ms
  • W ithout synchronization events occurring!
  • PCAP seems to be timestamping at the

application level.

slide-13
SLIDE 13

13

Future work

  • Evaluation of tools in real-time while in use.

Get a accuracy value for the collected trace, upon completion of the trace.

  • H/W implementation of a reference traffic

generator.

slide-14
SLIDE 14

14

A Method to Estimate the Timestamp Accuracy of Measurement Hardware and Software Tools

Questions?

Patrik.Arlos@bth.se Blekinge Institute of Technology