Cure for stats and interleaving OCD
I thought it might be of interest to report the results of my recent speed capping tests.
The CAB is ECI, modem HG612 and the ISP TalkTalk Business who are unclear over whether they provision on stable or standard profiles, I have been advised of both!. Definitely not the fast profile. I believe the profile is stable as DLM hits have been 100% consistent with stable. Interleaving has been produced about 7 times (once for no obvious reason) and always by odd events which mean that when it is applied it often would not otherwise be re-invoked for another month or so!! The line has always recovered fast path.
I am able to cap the speed down with the transmitted power remaining fixed. However when the upstream speed is capped the transmit power can also drops. More annoying is that a cap from 20 to 15 or 10 gives no benefit but a SNRM of 6 and large power drops. Maximums in the upstream SNRM and benefit occur at caps of about 16Mb/s and 11Mb/s
With no speed cap the line runs at about 71Mb/s down and just makes 20 Mb/s up. Should interleaving be applied the speeds become 63Mb/s and 18Mb/s.
The average error rates Downstream on fast path are:
No speed cap 71Mb/s ES/hour ~20 CRC/ES ~10 SES/hour ~0.5 SNRM ~6db
Capped to 64Mb/s ES/hour ~10 CRC/ES ~5 SES/hour ~0.2 SNRM ~9db
Capped to 56Mb/s ES/hour ~2 CRC/ES ~1.5 SES/hour ~0.0 SNRM ~12 db
capped to 48Mb/s ES/hour ~0.5 CRC/ES ~1.0 SES/hour ~0.0 SNRM ~15 db
The line suffers from odd error bursts and in spite of the acceptable average error rates, With no speed cap, it seems that roughly once month the ES limit of 1440/day or 720/day is exceeded. With the speed cap of 64 the ES limit has been exceed twice in 18 months. I suspect the ES limit won't ever be exceeded with a cap of 56 or less. The ratio of CRC/ES shows that many of the ES when uncapped are due to large bursts of CRC. If I use a 63168 chipset based modem these error rates go up by a factor of 2 or 3, the increase in errors is not uniform hour by hour but rather the 63168 chipset seems to respond very badly just to the large bursts of CRC and gives overall CRC/ES rates of about 30.
The average error rates Upstream on fast path are:
No speed cap 20Mb/s ES/hour ~3 CRC/ES ~1.1 SNRm ~6db
Capped to 16Mb/s ES/hour ~1.0 CRC/ES ~1 SNRM ~9db
capped to 11Mb/s ES/hour ~0.1 CRC/ES ~1 SNRM ~12db
Upstream does not seem to seem to have the same odd error bursts as downstream but has had instead steady high CRC rates of 1-5 CRC/min for several hours some evenings. These events have got very close to error limits. So far this has not caused a DLM response but some evenings I have powered the modem down for the night when the upstream daily count has exceeded about 500ES.
My noisy line looks as though it needs to be capped to something like 56/12 or less to be close to 100% unlikely to ever be hit by an error burst and then go interleaved the next day. With a higher speed cap of 64/16 interleaving events are not common but as noted have occurred. If I change ISP and have a fast DLM profile then either a much more modest speed cap or no speed cap should be OK.
The good news is that whilst I notice interleaving I do not notice any impact of these speed caps --- except of course in a speed test! (I am of course fortunate to start with 71/20 uncapped.) On a line like mine totally getting rid of any possibility of interleaving seems to cost up to 25% of the sync. Interleaving itself would cost 10% of the sync plus higher pings. Running at the speed that interleaving would give. i.e. roughly 64/16 in my case makes interleaving unlikely but not impossible.
I have made this post in case anyone on an ECI with no G.INP and who hates interleaving would be interested in an example of the exact impacts of capping.