On openbsc master, in do_lchan_free(), a T3111 timer is started in case the lchan is in an error state. Why?
3GPP TS 04.08: The sole purpose of timer T3111 is to let some time [pass] to acknowledge the disconnection and to protect the channel in case of loss of the acknowledge frame.
I'm asking because for dyn PDCH, I'm looking at jolly's branch, which adds channel defragmentation: when a channel is released, do_pdch_defrag() closes gaps in the assignments by swapping active channels into empty slots. It appears that the defragmentation code doesn't want to wait for T3111 ... and prefers to run through channel reallocations more quickly, if that makes any sense?
Also, I'm not sure whether we want the defragmentation feature at this time. It seems to be orthogonal, not related to dynamic reassignment of TCH <-> PDCH per se.
Thanks for any opinions or hints!
For details, take a look at a6c1c652e5f9cd7f2e456af39a54b8fafdc5b344 openbsc/src/libbsc/abis_rsl.c:69:do_lchan_free() (branch users/neels/dyn_pdch)
~Neels
On 23 May 2016, at 18:46, Neels Hofmeyr nhofmeyr@sysmocom.de wrote:
On openbsc master, in do_lchan_free(), a T3111 timer is started in case the lchan is in an error state. Why?
we received a connection failure indication (e.g. LAPDm retransmissions missed N200+1(?) times with T200 for the timeout). Now just because the BTS gave up does not mean the MS stopped broadcasting. It might continue to send it SACCH and cause interference. So we wait a bit before re-using the channel.
Also, I'm not sure whether we want the defragmentation feature at this time. It seems to be orthogonal, not related to dynamic reassignment of TCH <-> PDCH per se.
I just guess. For GPRS some channel combinations are better for multi-slot assignments than others. So the defrag can be beneficial for some MS classes and some configurations. I don't remember out of my head which classes and which patterns are wanted.
holger
On Mon, May 23, 2016 at 07:15:59PM +0200, Holger Freyther wrote:
I just guess. For GPRS some channel combinations are better for multi-slot assignments than others. So the defrag can be beneficial for some MS classes and some configurations. I don't remember out of my head which classes and which patterns are wanted.
Sounds quite limited, and also the patch has two potential problems, so I will postpone the defrag part.
~Neels