Change the counters_store_db function to be a generic for_each
function taking a function pointer and data. Use that in bsc_hack
to store it to the DB.
This is removing the DB requirement and will allow to handle
the counter values in different ways without making the counter
list public.
I verified that the syncing is still taking place.
This is the new logging architecture, including
* support for multiuple logging targets like stderr and vty
* log levels in addition to categories/subsystems
* filtering based on imsi, i.e. only see events for one subscriber
* dynamically change log level for each category for each vty
This has the advantage that counters can be added all over the code
very easily, while having only one routine that stores all of the
current counter values to the database. The counters are synced
every 60 seconds, providing relatively fine grained statistics
about the network usage as time passes by.
Tweaking theses can be useful especially tx-integer that influence
both the spread of rach attemps and the delay between two attemps.
Looking up GSM 04.08 3.3.1.1.2 & 10.5.2.29 can help determine good
values. The default are choosed with a wide spacing between attemps
(tx integer = 9 -> T=12 & S=217 (non-combined CCCH/SDCCH) or 115 (for
combined CCCH/SDCCH)). This alleviates the problem of responding to
several RACH attempts by a same MS, allocating several RF channels when
only 1 is needed.
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
The previous implementation had some shortcomings:
- If the MIN ID given was not the exact id of the first unsent SMS,
it would try to submit the same sms several time until id++ finally
made id go to the next one.
- If a subscriber had several SMS pending it would try to submit
them individually (only to get rejected because a paging for that
subscriber was already in progress)
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
we need to set newbfd->priv_nr to 2+trx_id, rather than keeping
it '2' all the time, as it is used to look-up the e1i_ts when
we receive a packet. A constant '2' would always match to TRX 0.
we also need to keep one separate bit for each TRX state in order
to properly generate the EVT_E1_TEI_UP event for trx > 0.
This implements the handover algorithm (and associated parameters)
as described in Chapter 8 of the book "Performance Enhancements in
a Frequency |Hopping GSM Network" by Thomas Toftegard Nielsen and Jeroen
Wigard.
The parameters such as averaging windows are configured in struct
gsm_network. We keep some state to trakc up to 10 neighbors as
they are being reported from the MS.
This has so far only been tested in a network with two BTS that
have each other as neighbor. Networks with morge neighbors might
encounter bugs.
This allows us to block packets that we have received after the channel
is no longer being used. This is visible during handover, where we still
receive a measurement report after the MS has switched to the new channel.
This leftover measurement report then attempts to trigger another handover,
which si bogus and will fail - and thus only consumes resources.
With the new LCHAN_S_ACTIVE state, we can check for this when processing
the measurement report.
This provides two functions: get_meas_rep_avg() to obtain the sliding
window average of one particular field, and meas_rep_n_out_of_m_be()
to check if at least N out of M measurments are >= BE.
If a RF channel is assigned but no response is ever heard from
the phone, we will receive a CONNECTION FAIL from the BTS,
triggering a RF release freeing the channel. Then sometime later,
T3101 will expire as well and free the channel again ...
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
- Need to use sms.id for the ORDER BY since 'subscriber' also has 'id'
- Need to add the join clause between 'SMS' and 'subscriber'
- Add a LIMIT 1 (probably no impact for the db size we're dealing with
here, but with large DB and mysql/postgresql this can help the planner)
- (fix a wrong comment in passing ...)
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
This patch takes care of handling the RTP streams / sockets during
an in-call handover from one BTS to another BTS.
It only works in combination with rtp_proxy mode.
This is not really nice, but we will soon have multiple users of
the CRCX / MDCX / DLCX signals, and we cannot guarantee the ordering
of them. So as a workaround, we move the RTP socket creation and
deletion into the core abis_rsl codebase.
We cannot support in-call handover of calls without a RTP proxy,
since at the time of the handover the SSRC, sequence number and
timestamp of the RTP frames change.
Since the MNCC API can now send and receive frames to/from the MNCC
application, we can also implement a proxy this way. Not at the RTP/UDP packet
level, but at the 'TCH speech frame' level.
Especially for handover, we need this mode as the receiver in the BTS needs a
persistent SSRC and monotonic frame numbers / timestamps.
During handover, we will not send RTP frames for quite some time. However,
the way the rtp_send code is structured, it will increment the timestamp
with a fixed amount every time we send a frame, independent how much wallclock
time has actually passed.
This code is a hack to update the sequence number and timestamp in case it
seems to be wrong. It makes handover much more reliable.
Instead of passing TRAU frames down the MNCC API to the call control
application like MNCC, we now decode the TRAU frame into the actual codec
frame. We do the same with the RTP packets in case of ip.access and
thus have a unified format of passing codec data from the BTS to
an application, independent of the BTS type.
This is only implemented for V1 full-rate at the moment, and needs
to be fixed.