Currently an existing DL TBF can get lost in the process of
establishing an UL TBF via RACH. This can lead to stalled connections
until the network sends more LLC frames.
This commit adds a check for a non-empty LLC queue after the UL TBF
has been established to rcv_control_ack (GPRS_RLCMAC_UL_ASS_WAIT_ACK
path) to eventually establish a new DL TBF on the UL TBF's PACCH.
Sponsored-by: On-Waves ehf
Currently an existing DL TBF is freed immediately, when a resource
request is received. This makes sense since the MS might have dropped
it when switching to the PDCH signaled via the AGCH for the SBA. But
if the TBF still is assumed to exist on the MS side, there might be
TFI collisions if the old TBF object is not kept to block its TFI
for some time.
This commit changes rcv_resource_request to call release() instead of
tbf_free() on the DL TBF object (if it exists).
Sponsored-by: On-Waves ehf
Currently reuse_tbf (partly) resets the old DL TBF and uses its PACCH
to establish a new DL TBF. The method can not be used with UL TBFs.
This commit replaces the reuse_tbf method into a
gprs_rlcmac_dl_tbf:release method which triggers the TBF's timer
based deletion (so that the TFI is still reserved for some time) and
a gprs_rlcmac_tbf::establish_dl_tbf_on_pacch which can establish DL
TBFs on existing PACCHs of either DL or UL TBFs.
Sponsored-by: On-Waves ehf
Currently when a second MS object has been created for an MS, because
the TLLI was not known yet, the will be detected in
gprs_rlcmac_tbf::extract_tlli and the two objects will be merged by
update_ms. But when the dl_tbf is moved from the old to the new
(second) MS object, the old MS object can get idle and be removed
before the object are merged. This can cause LLC frame loss when the
MS object is deleted immediately after getting idle (no timeout
configured).
This commit adds a guard to keep the MS object until extract_tlli has
been executed.
Sponsored-by: On-Waves ehf
This reverts commit e91bd3babd.
That commit seems to cause hanging DL TBFs when there was a RACH
based UL TBF establishment while it that TBF is active. This could be
caused by the use of a different PDCH for the SBA.
Conflicts:
tests/tbf/TbfTest.cpp
tests/tbf/TbfTest.err
Currently all of these messages are discarded if they are assumend to
be caused by noise. But even in these cases, the FN and TN values
which are added by the DSP are valid. So these can be used to update
the current_frame value.
The osmo-bts sets the fBFILevel of a physical channel to -200dB if it
is used for PDTCH or PACCH which is the case for all PDCH. This way
a data_ind or ra_ind message is already send at least once per block
period (4 frames) per PDCH. These messages are passed to either
handle_ph_data_ind or handle_ph_ra_ind even if they contain garbage
data.
The ra_ind messages are sometimes sent a few frames earlier than
data_ind messages using the same frame.
This commit adds calls to update the current_frame value based on all
of these messages before they are discarded. The FN taken from ra_ind
are passed with an increased max_delay (5) to compensate for early
ra_ind messages.
Sponsored-by: On-Waves ehf
Currently a log entry is written if FN_data_ind - FN_time_ind <= -13.
This commit adds a counter 'rlc.late-block' that is incremented in
these cases.
Sponsored-by: On-Waves ehf
The FN of the data_ind taken from the DSP are monotonic, so latency
does not affect the detection of poll timeouts if these FN are used.
If the FN is larger than a poll_fn value, it can safely be assumed
that the poll response will not arrive later on.
Currently a max_delay of 60 frames is used, which has the drawback
that additional ~250ms will pass until a lost ACK is detected.
Using the data_ind's FN alone breaks the poll timeout detection if
there are no other MS sending data blocks.
This commit adds BTS::set_current_block_frame_number that is called
with the FN taken from data_ind messages. The max_delay is set to 0
which removes the additional delay, when this FN is used to detect
poll timeouts. So the average additional delay decreases with the
number of data_ind per time. The current_frame is updated unless it
seems to have been updated already (assumed if 0 < cur_fn - block_fn
< 500). Thus the time_ind has still priority to update the
current_frame value.
Sponsored-by: On-Waves ehf
Currently the max_delay parameter is set to 13, since that is
slightly above maximum number of frames that a time_ind can preceed a
block's data_ind of the same frame. This assumes that these messages
are not reordered after thay have been obtained from the DSP. In the
current implementation, the GPRS data_ind can directly be taken from
the DSP by the PCU while the time_ind messages are provided via the
BTS. So the messages are queued differently in that case, resulting
in a additional delay of the data_ind with respect to the time_ind.
The propability for this raises with a increased CPU load of the PCU.
If this happens, a poll timeout is detected by mistake and the poll
is either retried or cleared.
This commit increases the tolerance to 60 frames, since
values for FN_data_ind - FN_time_ind of up to 50 frames have been
observed under heavy PCU load.
Sponsored-by: On-Waves ehf
Currently the maximum additional delay is hard coded to 13. This
value depends on the source of the frame number value.
This commit adds the max_delay parameter to make it caller dependant.
Sponsored-by: On-Waves ehf
This commit adds the relevant frame number to the "poll timeout"
logging message. In addition, logging is added to the places where
poll_fn gets set.
The goal is to track down the source for frequent "poll timeout"
messages.
Sponsored-by: On-Waves ehf
Currently when allocating tbf_alloc_ul_tbf or tbf_alloc_dl_tbf
objects, the allocated memory area is pre-initialised by talloc_zero
before the C++ constructors are called. This is not recognised by
Coverity, since there is no talloc model file yet. Thus Coverity
complains about missing initialisers.
On the other hand, it is still planned to convert the TBF classes
into real C++ ones. So instead of silencing Coverity directly, this
is an opportunity to do it the C++ way.
This commit adds initialisers and initialisation code for all
members that relied on talloc_zero. The corresponding calls to
talloc_zero are replaced by calls to talloc to give ASAN/valgrind
a chance to detect future initialisation errors. Some initialisation
code is also moved from setup_tbf to the constructors, notably the
initialisation of the bts pointer.
Fixes: Coverity CID 1320604, 1320605, 1320606
Sponsored-by: On-Waves ehf
Currently the per IMSI/TLLI view only shows the number of packets and
the 'all' view does not show any LLC related information at all. A
constant LLC queue length is often an indication for a stalled TCP
connection where the RLC layer has stopped to send downlink data
messages.
This commit adds the number of packets to the 'all' view and the
number of octets to the IMSI/TLLI views.
Sponsored-by: On-Waves ehf
Currently when calling GprsMs::attach_tbf and a TBF of the same
direction already exists, the old TBF gets detached from the MS
object.
Therefore that TBF object loses access to that MS object including
for instance TLLI and IMSI.
This leads to failing DL TBF reuses, since the downlink assigment
cannot be sent on the PACCH later on because that must be sent on the
old DL TBF which ms() is NULL and the new DL TBF cannot be retrieved.
This commit fixes this bug by changing the GprsMs implementation to
keep a list of replaced (old) TBFs. TBFs are only removed when they
are being detached explicitely (see tbf_free and set_ms).
Addresses:
tbf.cpp:741 We have a schedule for downlink assignment at uplink
TBF(TFI=1 TLLI=0xf35a680e DIR=UL STATE=RELEASING), but there is no
downlink TBF
Sponsored-by: On-Waves ehf
While this does not happen in real use, and unset btcx can lead to
segfaults in test cases. The other code outside of gprs_bssgp_pcu.cpp
does not depend on bctx being non-NULL:
Sponsored-by: On-Waves ehf
The missing const qualifier prevents the llist_empty() C++ wrapper
function from being compiled successfully when it is used.
Sponsored-by: On-Waves ehf
The TBF object are currently created by using talloc_zero/talloc_free
directly from plain functions. Therefore C++ constructors and destructors
are not called. So the only initialisation that is done is setting
every member to 0. Non POD members do not have their constructors
called either, which makes it impossible to use the current LListHead
class for real members when the LListHead::m_back member has to be set.
This commit changes the TBF allocation functions to call the
corresponding C++ constructor after the call to talloc_zero and to
register the C++ destructor with the talloc context, so that is is
called before talloc_free actually frees the memory.
With this change, non-POD members and custom
constructors/desctructors can be used with gprs_rlcmac_tbf,
gprs_rlcmac_dl_tbf, and gprs_rlcmac_ul_tbf.
Note that this change is only a single step of the plan to turn the
TBF classes into real C++ classes.
Sponsored-by: On-Waves ehf
This tests the usage of an existing TBF that is no longer in FLOW
state to request a new DL TBF via the old TBF's PACCH.
The test triggers a bug that breaks the association between both TBF
objects, resulting in packet loss and transmission stalling.
Sponsored-by: On-Waves ehf
Currently the assignment is completed by manipulating the state of
the TBF objects directly by setting the state fields to fixed values.
This way, the PCU's code that is responsible to update the state
accordingly is not tested.
This commit changes this to simulate RLC Control Acknowledgement
messages instead.
Sponsored-by: On-Waves ehf
This function basically request the generation of the next downlink
RLC block. Since this will no really send somthing to the PCU, the
current name can be misleading.
This commit just renames the function.
Sponsored-by: On-Waves ehf
The block number can always be deduced from the frame number. The
current test code handles the block number explicitely, which makes
the code more complex and has also led to block number errors cause
by not wrapping the numbers (valid block numbers range from 0 to 11).
This commit changes send_rlc_block to always compute the block number
based on the frame number. It also turns the block_nr into an
optionaly output-only parameter.
Sponsored-by: On-Waves ehf
The current implementation takes a lot of parameters (bts, trx_no,
...) that can also be taken from a TBF object.
This commit adds an alternative variant with just takes a TBF, the fn
(in/out), and the block number (in/out).
Sponsored-by: On-Waves ehf
Currently the pending LLC packets are lost in some cases when MS
objects are merged, for instance after a RACH when there were 2 MS
object for the same MS (they get merged, when the TLLI is known for
both objects).
This patch modifies GprsMs::merge_old_ms to move all pending LLC
packets (if there are any) to the current MS object.
Sponsored-by: On-Waves ehf
This methods takes all LLC frames from the old LLC queue and moves
them into the current. If both queues are ordered chronologically
(recv_time), the resulting queue is also ordered.
Sponsored-by: On-Waves ehf
Currently all active TBF of an MS are killed if a Packet Resource
Request is received from the MS. In general this happens after a RACH
request. This does not happen after a resource request that has been
included into a Downlink Ack/Nack.
Sometimes an UL TBF is requested by an MS via RACH while a DL TBF is
running for instance to send a TCP Ack. This can happen, if a former
request via PACCH did not work.
This commit removes the killing of the DL TBF from
gprs_rlcmac_pdch::rcv_resource_request().
Sponsored-by: On-Waves ehf
This adds tests for
- RA update with RACH for the RAUpdateComplete message
- RACH for UL while DL is active (LLC queue not empty)
Sponsored-by: On-Waves ehf
This commit fixes several issues:
- Set MS class in request
- Set IMSI in establish_ul_tbf_two_phase
- Fake assigment acknowledgement in establish_ul_tbf_two_phase
- Fix TFI bit offset to 1 (was 2)
Sponsored-by: On-Waves ehf
Currently the functions test_tbf_single_phase and test_tbf_two_phase
do the test logging, BTS intialisation, and the complete message
sequencing on their own. Therefore they cannot be used to test more
complex sequences like TBF reestablishment.
This commit moves the code that does the actual messaging into own
functions. The frame number handling is generalised which also fixes
a block number wrapping error on the way.
Sponsored-by: On-Waves ehf
Currently the merging of the meta information (MS class, IMSI) takes
place in gprs_rlcmac_tbf::merge_and_clear_ms(). This makes it
difficult to merge the internal state and does not directly relate to
TBFs anyway.
This commit moves this into a new method GprsMs::merge_old_ms.
Sponsored-by: On-Waves ehf
Currently the signal S_NS_ALIVE_EXP emitted by the NS layer if the
alive check has timed out too often is ignored. This prevents the PCU
from reconnecting to the SGSN if it has not been accessible for some
time.
This commit modifies nsvc_signal_cb to reset the NSCV if
S_NS_ALIVE_EXP is sent, so that the PCU continues to send NS RESET
message if that happened.
Sponsored-by: On-Waves ehf
Currently there is not support for Network Service (NS) logging.
This commit adds the missing definitions and sets the default level
to INFO. Further configuration can now be done with the 'logging
level ns' VTY command.
Sponsored-by: On-Waves ehf
This commit fixes the go_parent_cb API according to libosmocore's
commit of the same name.
Fixes:
pcu_vty.c:799:2: warning: initialization from incompatible pointer
type [enabled by default]
.go_parent_cb = pcu_vty_go_parent,
Sponsored-by: On-Waves ehf
The IMSI '000' is used as default value for an incoming BSSGP
message's IMSI IE. This can lead to the retrieval of the wrong MS
object from the storage.
This commit changes the get_ms method to skip the IMSI search if such
an IMSI is passed as selector.
Note that changing the default value in the BSSGP code does not help
here.
Sponsored-by: On-Waves ehf
The confirm_tlli method does not handle TLLI clashes in the MS
storage.
This commit changes gprs_rlcmac_dl_tbf::handle() to use update_ms
instead.
Sponsored-by: On-Waves ehf
Currently if an MS retries to access the PCU by using RACH and if
there is already an entry for that MS, a duplicated MS object
referring to the same TLLI is created. This is caused by blindly
setting the TLLI without querying the MS storage to avoid
inconsitencies.
This leads to several entries in the MS storage that are assigned to
the same TLLI. If that happens, 'show ms all' can display multiple
entries with the same TLLI (note that an MS object can belong to
several TLLIs, so there might be an intersection that is not visible
in the list) or 'show tbf all' can show entries with MS_CLASS == 0 in
some cases.
This commit changes update_ms() to merge and clean up old entries
that belong to the given TLLI if they exist. Some data (like the MS
class) is copied to the new MS object.
Note that TBF belonging to the old MS object are deleted immediately
if they have not registered a timer.
Sponsored-by: On-Waves ehf
This commits sets the initial CS to 2 to allow a successful
connection setup if the radio link has a low quality. The slot
allocation algorithm is changed to 'dynamic', which is the binary's
current default anyway.
Sponsored-by: On-Waves ehf
The current default interval is 2s which seems to be too short when
the DL TBF has to be established. This may cause freezing or really
slow TCP connections.
This commit increases the default value to 4s. When the
dl-tbf-idle-time is set, DL TBF are established less frequent, so
smaller values (like 2s or below) can be used to improve the average
latency when the load is high.
Sponsored-by: On-Waves ehf
When doing an RA Update the network can request to change the TLLI.
In this case, there can be 2 MS objects with different TLLI for a
single real MS. The first is associated with the old TLLI and the
IMSI, while the second is associated with the new TLLI and no IMSI if
it had been created for the uplink TBF. When the first message with
the new TLLI and the IMSI arrives from the network, the PCU is able
to detect this.
Currently this is not handled properly. The TBFs of the old MS object
are not cleaned up properly, keeping the old MS from being deleted.
This patch modifies gprs_rlcmac_dl_tbf::handle to check for this and
if neccessary to move an existing DL TBF and to clean up the old MS
object to ensure its deletion.
Sponsored-by: On-Waves ehf
Currently when receiving a PACKET DL ACK/NACK message with the Final
Ack Indicator bit set, the TBF's state is set to
GPRS_RLCMAC_WAIT_RELEASE but T3193 is only started when the LLC queue is
empty. Otherwise the reuse_tbf() method is called to establish a new
DL TBF. In that case, the timer is not started. This will leave the
current TBF without a timer so it is potentially not released later
on.
This is recognisable by sticky entries in the output of the
'show tbf all' command and possibly allocation failures if there are
too many of them.
This commit changes the code to always start T3193 to make sure, that
a timer is always active when the the state is set to
GPRS_RLCMAC_WAIT_RELEASE.
Note that TS 44.060, 9.3.2.6 requests to release the 'old' TBF
immediately in some cases, which is not implemented by this change.
This will lead to a longer reservation period of the TFI only, which
is safer than reassigning it too early.
Sponsored-by: On-Waves ehf
Currently the initial_cs_dl value is used to compute the maximum leak
rate. This can be too low if adaptive CS selection is used.
This commit changes gprs_bssgp_tx_fc_bvc to derive the max CS level
from the configuration.
Sponsored-by: On-Waves ehf
Currently CoDel is disabled by default.
This commit enables CoDel on start up with the default interval time,
equivalent to the 'queue codel' VTY command.
To disable CoDel, use the 'no queue codel' command.
Sponsored-by: On-Waves ehf
Currently packets are only dropped if they have reached their maximum
life time. This leads to LLC queues being constantly filled under
load, increasing the latency up to the maximum life time. This kind
of bufferbloat hinders TCP's congestion avoidance algorithms. To keep
the queues short, the CoDel active queue management algorithm can be
used.
This commit changes to llc_dequeue method to apply the CoDel
algorithm to selectively drop LLC frames before they passed to the
TBF layer to be encoded in BSNs. This feature is currently disabled
by default.
The CoDel state is managed per MS since the LLC queues are also kept
in the MS objects.
Note that there is still some buffering in the TBF objects, in the
worst case (CS4) 3.5kByte + LLC-MTU octets are stored there. The
resulting additional packet delay is not (yet) taken into account for
CoDel.
Also note that configuration changes are applied to new MS objects
only.
The following VTY commands are added to the 'pcu' node:
- queue codel activates CoDel, the interval is selected by
the implementation
- queue codel interval <1-1000>
activates CoDel with a fixed interval given
in centiseconds (10ms-10s)
- no queue codel deactivates CoDel
Which interval value to use is still an open issue. For high speed
links (e.g. Ethernet), CoDel suggests 100ms. For slower links, the
expected RTT is recommended. The current implementation uses a
default value of 2000ms.
Measurements:
Note that the following measurements depend on several other factors,
most notably the interaction with the SGSN's flow control. They are
just examples to give an idea how CoDel might influence some
parameters.
The measurements have been done with a single E71, first with a
running ping only (Idle), then with an additional TCP download
of a 360k file (Busy). The CoDel interval was set to 1s.
- Idle :
ping ~400ms, avg queue delay 0ms, dropped 0
- Busy, No CoDel:
ping ~6s, avg queue delay 4-6s,
dropped 0, scheduled 948, duration 54s
- Busy, CoDel:
ping 500-1500ms, avg queue delay ~600ms,
dropped 77, scheduled 1040, duration 60s
More measurements with two MS downloading in parallel (two
independant measurements per case).
- Busy, No CoDel:
dropped 0, scheduled 1883, duration 121s
dropped 19, scheduled 2003, duration 133s
- Busy, CoDel:
dropped 22, scheduled 1926, duration 116s
dropped 22, scheduled 1955, duration 108s
Sponsored-by: On-Waves ehf
This commit adds an implementation of the CoDel algorithm based on
the reference pseudocode presented in
http://queue.acm.org/appendices/codel.html. Instead of abstracting
the queue itself, the implementation provides a time stamp based
automaton which is invoked after a package has been dequeued.
Note that the modifications of the algorithm shown in
https://tools.ietf.org/html/draft-ietf-aqm-codel-01 are not yet
applied.
Sponsored-by: On-Waves ehf
Currently the values Bmax/R default MS are computed under the
assumption than min(4, N_PDCH) DL slots are allocated for an MS, even
if multislot assignment is not enabled.
This commit changes the computation to assume 1 DL slot if algorithm
A is selected or the dynamic algorithm is used and has disabled
multislot assigment due to high load.
Sponsored-by: On-Waves ehf
Jacob Erlbeck