BenYeeHua
Yo man you are back again.
If anyone here don't want crippling performance, just stick to Google DNS. Every night your DNS getting cripplied by congestion is no fun.

Looking at your trace, I think you might be the first documented case of last mile congestion.

Since you said upload speed is okay, this aligned with the usage pattern during the evening where people mostly just consume content aka download.

You can easily monitor this because it will be a daily occurrence.

There's nothing you can do to fix this. There's probably nothing a TM staff attending your case can do either.

The solution to last mile congestion is XGSPON / 25GPON / 50GPON and a faster ring network. But according to all the MBA in TM, we don't want fast internet.

If your place have Allo, can give it a try. Reason is, I don't know what kind of congestion you are having. TM have congestion from last mile all the way to the international gateway.

With Allo, it is totally different network from last mile all the way to international gateway, so you cannot possibly hit the same congestion issue.

With other telco on TM infra, you will have the CoS problem with OLT port too. This is where TM only have 2.5gig of bandwidth per port to sell. What they then do is have a Class of Service so higher speed tier get more priority. So if you subscribe to shit tier like 30mbps or 100mbps, you are gonna get throttled badly at the port level itself. If you are like me at top tier 2Gbps, I basically cut queue and have priority access. This is how TM offer their Unifi Ultra plan. The marketing material says they are selling the speed. In reality, they are selling priority service. The port is 2.5gig for many houses to share.

EDIT:
I did follow along in Allo thread just reading stuff but never bother replying. From what I try to understand based on people's description, most of the network downtime is at the ODN / OLT, which is totally different and isolated for different places.

I also suspect it is just one or 2 uplink from the OLT. They might not use a ring network like TM.
Reason being, if they use a ring network, they should have really high uptime.

But all these doesn't matter as long as they don't oversell capacity. Plus they are cheap. When it goes down just stick to your mobile data. I don't believe they are so horrible to go down once a month.

This post has been edited by kwss: Aug 21 2025, 10:30 AM

Just sharing what is a ring network. If you just google this term you get a lot of kiddo or dinosaur answer.

High level explanation from Huawei:
High-Reliability Ring Network
https://info.support.huawei.com/hedex/api/p...2004996294.html

Detailed setup from Cisco:
ITU-T G.8032 Ethernet Ring Protection Switching
https://www.cisco.com/c/en/us/td/docs/wirel...ter_0111111.pdf

Components of the ring from Juniper:
High Availability User Guide: Understanding Ethernet Ring Protection Switching
https://www.juniper.net/documentation/us/en...erstanding.html

Finally, the actual specification:
ITU-T G.8032/Y.1344
https://www.itu.int/rec/dologin_pub.asp?lan...DF-E&type=items

Network with high uptime don't just drop from the sky. There are a lot of serious engineering and money thrown into it.

The problem with Malaysia GLC companies are once the thing is built, there is no continuous investment to update the infrastructure.
Example:
1. KLIA core switch was 25 years old when it finally died and the whole airport grind to a halt.
2. KLIA aerotrain.
3. RapidKL train, bus, lift and escalator.

I am sure DNB will be in the same situation 20 years later.

This post has been edited by kwss: Aug 21 2025, 11:26 AM

The biggest problem is congestion, not really "routing problem".
TM has way too many IX peering to faraway places. With big tech and cloud provider, 75-80% of traffic will be very CDN and edge heavy.

TM then had a problem where their connection to MyIX, Singapore got overloaded while many of their other faraway IX connection remains unused.

They then try to "load balance" them to all the other connection, effectively jacking up latency. Simple route that should just end up in Equinix SG will now traverse transit provider.

There's two way TM can fix this.
1. Follow Maxis footstep and cut the number of IX peering to minimum. This method does affect their CAIDA score.
2. Follow Singtel and be just private peer with everyone. Then throw money into having huge capacity, like multi Tbps local peering capacity. This way they can keep their CAIDA score high while avoiding congestion.

Now CAIDA score is just some ego ranking. It has no relevance whatsoever to user experience. However, it does indicate network diversity. What does that do to an eyeball network? Nothing. Lots of work for no real benefit.

Eyeball network just need multiple transit provider as diversity.

To be fair if you go out and randomly ask people which one is a better network; the one with big pipe to 2 IX, or one with many small-medium pipe to 16 IX.
Generally people will choose the 16 IX. It's a logical fallacy where if more is good, then keep jacking up the number is better right? After all it does improve your CAIDA ranking.

It's different if you are a Tier 1 or cloud / content provider. They will benefit by having local presence to do edge processing and also transit traffic to adjacent area.

But for TM, all the traffic must come back to Malaysia. TM said they want to do AI data center. But will their customer want to do edge computing in Japan, Korea, Singapore, UAE, London?

TM customer are almost all in Malaysia. Now they load balance everything towards different region, do you want your packet to come back from France? Hong Kong? Japan?

Then there's the average bandwidth graph which makes the one presenting it looks smart if they didn't understand the gotcha.

It will look like the connection is unused most days. When it's congested, the average usage is still within 90% or even 80%. You then make a conclusion link is not congested.

This is the same flaw as using MPLS-TE auto-bandwidth. The peak is actually 100%. The pipe blow up and traffic dropped dramatically. Rinse repeat. The final graph shows the average and people think everything is good.

Anyway, when HSBB first launch, everyone was on 5-20 Mbps. Now it's 30-800 Mbps. I just exclude Ultra plan since they said no one wants them.

With the same gear as many years ago, it won't fly anymore. TM is also using up excessive fiber optic cable in some busy last mile and just break up the ODN into different OLT port. It really is a lot of work for less and less return.

I can only conclude salary is still very low here. Laying more fiber and buying more GPON line card is actually cheaper than upgrading the gear.

You can do it till a point when your chassis / shelf uplink become the congestion point. Then the ring starts crawling. Then it become like KLIA aerotrain.

BenYeeHua PJng
I think this is a Microsoft bug. Like for Teams, once I get that error, I just open another tab and type the full signed in URL and it will just work, like
https://teams.microsoft.com/v2

I believe a lot of things are because management don't want to fix anything / change anything.
If you notice, TM product line is the same. The speed upgrade over the years is due to government policy.

Like MAB and Prasarana:
Use tax money to build infrastructure.
The price is not charity, isn't even that cheap.
When infrastructure fall apart, just do nothing and let it rot.
Wait for Menteri to show up and privately ask for tax money to repair / upgrade.

For public transport that affect the economy, maybe they get urgent attention. If not, they just buy a bus or ask you to walk like KLIA.
Evening congestion? LOL. Go sleep early or do something else.

Actually how they calculate the price?
1gig to 2gig = RM 20 diff
2gig to 5gig = RM 100 diff
5gig to 10gig = RM70 diff

We all know that nobody will ever get 10G on XGSPON. 8.5G is the max.
So how come a 3gig jump from 2G to 5G cost a RM100 premium?
But the slightly more 3.5gig jump from 5G to 8.5G only cost RM70.

If you say you get more discount when you buy more, then the 1G to 2G jump is still cheaper per gig.

If the NIC only have one hardware queue, the number of threads won't really improve things that much. The reason is only 1 CPU core will receive hardware interrupt. There's a reason why Mellanox NIC cost so much.

Try different Coalesce setting? Most consumer hardware need this setting to not wreck your CPU.

Astro Fiber is really just a rebadge of TM Unifi.
The last mile is from TM, the transit is also TM.
Astro is so lazy until they didn't even bother to connect to any Internet Exchange or multi-home.
Basically they are just reseller of TM Unifi.

You can verify my claims here:
https://bgpview.io/asn/56231

From TM perspective, the cost price should be lower than Astro.

Let me offer you another perspective.
Fiber cut upstream of the optical splitter will affect the whole area.
It's not possible the whole affected area has no Unifi user.
So in a way you will still get 24 hours repair.

Let's be honest, 99.7% uptime guarantee is very damn good for a home account. Just use mobile data in the meantime.

So this whole non-TM telco on TM infra == unreliable story is overblown in my opinion. Sometimes I feel it's made up just to discourage people from using other telco.

There are quite a few workaround if IPv4 is absolute must for you.
1. Use cloudflare tunnel
2. Sign up for a VPN that offers IPv4 complete with port forwarding. You get VPN to every region on earth + IPv4 with this method.
3. Use an OTT / Network as a Service that lease you anything via a tunnel. They are basically different kind of tunnel broker, some even offer BGP over GRE. If telco screw you with IP transit, this method will tell them to fly kite. There are companies that even offer IX peering over GRE. Yes, even to Equinix, worldwide!
4. Use a VPS and set it to forward IPv4 traffic. You get compute, storage, IPv4 with this method.

I don't see ISP offering IPv4 as an advantage today. Why pay RM10/month just for a number when you have better option?

Just so you know TM is lobbying very hard against multiple last mile provider.
I guess I could say it out loud since it is an open secret now.

I never receive this email and my IPTV VLAN is still intact.

But to answer your question, if the VLAN is gone, your TV box will no longer work.

This whole IPTV VLAN is a dumb idea from the start, inherited from enterprise networking.

Telco can achieve the same result purely with HQoS but then I suspect their China brand stuff cannot do it.

The wifi won't do anything for this tv box. It's only used for app you side load.

But for the white tv box, there's a streamyx mode which uses internet. However, since the email said the white box won't be working as well, I suspect they are changing vendor and hence a whole new system.

TM current IPTV and VOIP system are all Huawei.

Disabling IPv6 cannot solve congestion issue.
They will help if you refuse to use Google DNS, like using controld or nextdns.

There's no such thing as naik motor instead of driving in network.

Great to know this. I hope all telco who pull this stunt kena saman.

If you don't feel like spending money to upgrade anything, then watch this:


If you have devices at the far end of your house, which additionally means you have a hidden node problem:
https://en.wikipedia.org/wiki/Hidden_node_problem

Many people think they can buy a strongest signal AP to fix problem. No you don't. You trade one problem with another.

The problem is more complicated than that.
TM do not have enough capacity on their International Gateway. They have been dancing with this by using BGP to try and load balance traffic across different circuit.
Whenever Cloudflare reroute KUL traffic, their PNI is rendered useless and suddenly all those traffic flows through their IGW.
It's a congestion problem.
Everything in TM needs an upgrade.
Even their ring network do not have enough capacity in more and more places.

You all can try.
But I can tell you the government is having it easy with TM as they contribute quite a lot of money to the Budget.
Now with Petronas money drying up, all these companies are even more important to keep the Budget going.

It's in the government interest that TM do not spend any more money but continue to milk you more.