SUMMARY TO DATEThis post is to summarise some of the data and recap on the main findings & observations. It should be borne in mind that we have a lower representation of users on ECI cabs, but based on the statistics so far, I think its safe to say the following:
- Somewhere around 33%-40% of ECI lines have been upgraded so far.
It is probable that those who are on an ECI cab and who have had a DLM reset wont be updated again until the profiles are rolled out to the remaining lines first. In the case of the Huawei's BT tended to mop up the reset/new lines in batches every few weeks. Until then reset lines will be on the new default open profile which is INP=3 & Interleave delay=8 on the downstream.
- The majority of users have seen an uplift in sync speed - As expected, previously interleaved lines have benefited the most.
After what looked to be a disappointing start, more users are now seeing speed increases. For those lines which already sync at maximum, increases have been seen in the max attainable rate.
There are only 2 lines which appear to have lost a significant amount of actual sync speed. One of these has since had a new line (D-side) and DLM reset. Until DLM catches up with g.inp for this line, we are unable to draw any conclusions and I will be watching it closely. There are 2 lines currently syncing at max rate which appear to have lost some attainable speed.
- Lantiq based chipsets now appear to be getting better performance on ECI cabinets.
Pre g.inp, Broadcom based chipsets seemed to outperform the lantiqs when it came to connection speed. However, this seems to have changed since g.inp has been enabled on the ECI cabinets. The lack of full line stats on ECI modems and HH5A have in the past meant its difficult to monitor what is going on with these lines. However rp00 is using a hacked ECI modem which has the ability to look at line stats in greater depth. We are continuing to investigate this area, but several members have now proved much better performance on their lines with a Lantiq based modem since g.inp has been applied. From looking at the information available so far, it would appear that the Lantiq based modems have different framing parameters applied. These framing parameters are used to calculate how much over-head the line needs for Error Protection (RS/FEC). Less overhead means more available speed for sync. It has also been noted that whilst BCM based modems have Interleaved delay of 3 set on the bearer1 channel (which is used for g.inp). The ECI/Lantiqs do not. Investigation is still ongoing. But if the Lantiqs are set to have less RS overheads, then this will explain the better sync speeds.
- The 96.69% method of using IPprofile to check for g.inp does not work for VRX268 chipsets.
More specifically it would appear that this may be something to do with it being the ECI modem and HH5A not being able to support upstream g.inp. Both of these modems can support g.inp in the downstream direction only. For other lantiq based modem chipsets and those which have had firmware upgrades to support both up and downstream g.inp and the 96.69 v 96.79 calculation still appears to work. We do not know why this is, but it may have something to do with the different FEC overheads in the framing parameters set (see above). This is confirmed as applying to both ECI and Huawei cabinets.
Attached below is latest table for ECI g.inp line data.
- Those in the top block are all monitored by MDWS if you wish to look at any of the stats in greater depth.
- I have colour coded previous interleaved line data so that it is easier to spot which lines we would expect to achieve the best speed increases after g.inp has been applied.
Note that because of the way modems work out overheads for Interleaved lines - it is perfectly normal to for those lines to lose some attainable.
- Gains and losses have been cell coloured to easily spot positive or negative changes. Yellow indicates no change, or a change of between -100kbps to 100 kbps which is perfectly acceptable.
- Figures in red text are those with losses higher than anticipated.
Thank you to forum members and MDWS users for sharing their data with us.
