User Throughput versus Cell throughput

For Huawei 5G Performance Engineers:

I am having trouble with Huawei formula for DL user throughput and DL cell throughput.
Sometime avg DL user throughput is higher than DL avg cell’s throughput, especially in night time when there is low traffic. How is this possible?
Has anyone faced same issue?

1 Like

I hope this can help:

1 Like

But you are using a different formula than what Huawei recommends.
For DL User throughput.
And even in this case you get User Throughput higher than cell’s trhoughput.

Those counters of Huawei are pretty much tricky…

1 Like

Not sure about 5G, but I observed similar scenario in 4G due to difference in pegging nature of counters.

You can see similar situation here…

image

What i’ve heard is that one is counted at MAC layer another one at RLC layer.
So cannot be compared User and Throughput.

User level Throughput is always at RLC layer.
Because it has QCI information.

Cell Throughput is at MAC.

Who told this invalidate the question rather than providing reason. :slight_smile:
Layers just add few extra headers.
What if I remove them?
What will be the impact on time and bits being carried? :blush:

1 Like

Have you seen for other vendors that User Throughput is higher than Cell Throughput?

For Huawei vendor and in 4G, sometimes, the User Throughput is higher than Cell Throughput.
I am looking for the reason.

It depends on the counters.
User Throughput might be calculated with high precision counters (or other way around).

Sometimes it can happen because the User Throughput formula is not considering the last slot transfering time and volume.

So, consider a situation when the last slot data load was really significant. In situations like that, the Cell Throughput might be higher* than the User Throughput.

*For cells having a very large load, for almost all slots time interval, since the User Throughput considers just the time when traffic is carried.

This should no thappen because in User Throughput it is removed both time and volume for last slot.
Also Volumes of data with and without last slot are similar in order of magnitude (like every data would be sent in last slot).
Lot of weird things on Huawei counters.
Also can anyone share which is the counter that gives amounts of slots in which the scheduler worked for all users?
For example if in one hours (3600 seconds * 1000miliseconds * 2 slots for numerology1) there were 86% of slots used for scheduling( single or multi users scheduled) and in the rest of 14% slots no user was scheduled.
Which is that counter in 5G Huawei?
Which is the counter that gives the value of 86% in my example?
I know that there can be several numerology for different BWP. This is why Huawei time counters are in microseconds not in miliseconds.

Yeah. Both are not considered in the UserThroughput calculation.
This was implemented in order to avoid the majority of the cases where data load at the last time slot is not significant.
But in cases where the proportion of data at the last slot is bigger than the others (for any reason) it can happen.

Just to understand how weird are Huawei counters look at this graph for a single cell hourly for data volumes:

Volume of last slot is very high like scheduler is doing jobs in packs of 2-3 consecutive TTis instead of packs of consecutive 80-100 TTIs.

Same idea comes from this graph:

That scheduler works in very small packs of TTIs/slots.

For this cell here’s the avg number of users in UL and DL:

And I think it makes sens because TB size is huge in 5G like 300.000 bits or even more.

So scheduler doesn’t need 60-100 consecutive TTIs, it can serve any UE with any heavy download within 2-3 TTIs/slots and this is why last slot becomes very important in formula and we should not kick it out.

Are these graphs on cell level?
It gives us time info, but not number of SUbcarriers info.
Being cell level and consistent it’d mean interference profile would be consistent across the entire cell coverage.
Are users sparsely located or at the same spot?

Yes indeed singel cell, hourly.
No idea about users location.

My concern is that Huawei formula’s for KPIs are so weird and needs lot of digging and correlation to understand what exactly peggs every counter.

Agree.
Huawei kpis are usually very detailed but not easy to understand …

So you see, it makes no sense to remove last TTI from Throughput formula of user (neither the time part nor the volume part).
Because last TTI carry a lot of data.
TTI or slot.

Maybe it is correlated with the type of applications used at this cell.
In case of burst applications the use of smaller TTI numbers can be deployed.
It can demonstrate an efficient scheduler job.