It can take a while depending on the size of the document..please wait
Discuto
0 days left (ends 29 Aug)
description
Further info
LATEST ACTIVITY
LEVEL OF AGREEMENT
MOST DISCUSSED PARAGRAPHS
LATEST COMMENTS
MOST ACTIVE USERS
P52
Since the measurement client can only be guaranteed to know the TCP payload volume (i.e. the size of the file transferred), it would be necessary to calculate the number of packets required to transfer this TCP payload and then use this number to calculate the volume of TCP headers.
Add comment
P53
However the number of packets is a function of the TCP Maximum Segment Size (MSS), which is itself a function of the Maximum Transmission Unit size (MTU).
Add comment
P54
In addition to the above, the potential presence of TCP options introduces the possibility that the TCP header size is not fixed which further complicates the calculation.
Add comment
P55
The result of these factors is that it is impossible to accurately calculate the IP Payload volume from the TCP Payload volume. Therefore an adequate safety margin should be taken into account. Example calculations for the overhead are shown in the following tables for various sample values of MTU and TCP header size for both IPv4 & IPv6.
Add comment
P56
IPv4 (no IP options) | MTU | |
Average TCP Header Size | 1500 Octets | 1280 Octets |
20 Octets (no TCP options) | 1.37% | 1.61% |
40 Octets (Average 50% of max) | 2.78% | 3.28% |
60 Octets (Max TCP options) | 4.23% | 5.00% |
Add comment
P57
IPv6 (no IP options) | MTU | |
Average TCP Header Size | 1500 Octets | 1280 Octets |
20 Octets (no TCP options) | 1.39% | 1.64% |
40 Octets (Average 50% of max) | 2.82% | 3.34% |
60 Octets (Max TCP options) | 4.29% | 5.09% |
Add comment
P58
Note that these tables are intended to provide an illustration of the potential impact of these variables; however it’s expected that in practise the MTU will generally be very close to 1500 octets and the average TCP header size will be close to 20 bytes. Only the two percentage values highlighted in each table are considered relevant in most real world cases.
Add comment
P59
Therefore 3% TCP header overhead can be considered to include an adequate safety margin and it can be used in calculating the IP packet payload. The IP packet payload calculation is done by adding this 3 % value to the speed calculated based on TCP packet payload.
Add comment
P60
However it should be noted that BEREC considers that TCP payload volume is the most reliable one to use when calculating the upload/download speed.
Add comment
3.1.4Miscellaneous Details
P62
Both download and upload speeds should be measured in the same manner and reported in bits/second (e.g. Kbit/s or Mbit/s). Note that conversion factors between mega and kilo shall be base-10 rather than base-2 (i.e. 1KB = 1000 Bytes rather than 1024 bytes)
Add comment
P64
Where possible, it is recommended to mitigate the effect of the following inherent HTTP and TCP characteristics which could otherwise introduce error to speed measurements.
Add comment
P65
TCP connection speed limit: As the bandwidth of an individual TCP connection is limited to the bandwidth-delay product of the path in question, it is necessary to utilise multiple TCP connections in order to saturate that path.
Add comment
P66
HTTP considerations: Generally the use of HTTPS is recommended. When using plain HTTP, either the appropriate HTTP headers to prevent caching should be used, or unique URIs should be used.
Add comment
P67
In principle, any kind of IP packet could be used for latency measurements (e.g. ICMP, UDP or TCP). However the following considerations should be taken into consideration:
Add comment
P68
• Operating systems normally require administrator (root) privileges for sending ICMP packets. Also ICMP packets are often blocked by firewalls and antivirus software and hence they cannot be relied upon.
Did you know you can vote on comments? You can also reply directly to people's comments.