Feature #4520
open
gbproxy: Redundancy between NS-VCs (BSS Side)
Added by laforge over 4 years ago.
Updated over 3 years ago.
Description
Each osmo-gbproxy is interconnected via several E1 lines to the BSS.
The normal Gb interface load distribution functions must ensure that traffic is transparently passed only over those NS-VCs that are marked as UNBLOCKED, ALIVE.
- Related to Feature #4519: Load Sharing Function in osmo-gbproxy added
- Status changed from New to In Progress
- Assignee set to lynxis
- % Done changed from 0 to 80
this is already present in the wip-fr branches, but needs testing. See also #4519.
The NS_Emulation in TTCN3 also has load sharing code at this point, it has been manually verified to work.
There's a TC_load_sharing_dl()
which consistently passes here, at least when used with NS/FR to the [simulated] BSS.
As we now have automatic executing of the FR tests, we should make sure to enable it there.
This ticket remains open to
- make sure we have (and automatically execute) a related test for the NS/IP case
laforge wrote:
There's a TC_load_sharing_dl()
which consistently passes here, at least when used with NS/FR to the [simulated] BSS.
As we now have automatic executing of the FR tests, we should make sure to enable it there.
This ticket remains open to
- make sure we have (and automatically execute) a related test for the NS/IP case
So basically we need to execute TC_load_sharing_dl() also for the non-FR tests? Is there anything more to it?
load_sharing for udp has been implemented. The TTCN3 test case needs to be enabled and modified to work also with UDP.
TC_load_sharing_dl expects to have 4 NSVCs in the nsei, but the UDP test only has a single one.
Hi Daniel,
On Fri, Feb 19, 2021 at 01:09:45PM +0000, daniel [REDMINE] wrote:
This ticket remains open to
- make sure we have (and automatically execute) a related test for the NS/IP case
So basically we need to execute TC_load_sharing_dl() also for the non-FR tests? Is there anything more to it?
I think in FR all the weights are always equal, so we expect a (statistically) equal distribution
over all NS-VC within one NSE.
For IP with SNS, the weights can be differnt in two ways:
- some NS-VC can be data-only while others can be signalling only
- We should test with at least two of either to see that distributio within each class works
- NS-VCs within one class (signalling, data) can have different weight
- we should at least have some basic checks, i.e. that on average, twice the weight
leads to about twice the traffic compared to another NS-VC
Also available in: Atom
PDF