Attention is currently required from: neels, pespin.

View Change

4 comments:

• Patchset:

  • Patch Set #2:

    Hi all,
    I measured 20 minutes worth of performance data (in nsecs) from this and the other global BSC timers for comparison. This is on a BSC with 82 active BTS, 4500 SDCCH and 3200 TCH/H defined. During the measurement, 800-1000 SDCCH and 450-600 TCH/H were active at any given time.

     @hist_update_t3122_chan_load_timer:
     [4K, 8K)              13 |                                                    |
     [8K, 16K)             21 |@                                                   |
     [16K, 32K)            16 |                                                    |
     [32K, 64K)             2 |                                                    |
     [64K, 128K)            3 |                                                    |
     [128K, 256K)         331 |@@@@@@@@@@@@@@@@@@@@                                |
     [256K, 512K)         836 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
     [512K, 1M)            48 |@@                                                  |
     
     @hist_bsc_store_bts_uptime:
     [16K, 32K)            88 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
     [32K, 64K)            35 |@@@@@@@@@@@@@@@@@@@@                                |
     [64K, 128K)            4 |@@                                                  |
     
     @hist_bsc_store_bts_lchan_durations: 
     [4K, 8K)               9 |                                                    |
     [8K, 16K)             10 |                                                    |
     [16K, 32K)            31 |@                                                   |
     [32K, 64K)             2 |                                                    |
     [64K, 128K)            0 |                                                    |
     [128K, 256K)           6 |                                                    |
     [256K, 512K)        1208 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
     [512K, 1M)             2 |                                                    |
     [1M, 2M)               0 |                                                    |
     [2M, 4M)               2 |                                                    |

    I still have an accuracy problem to work out as it seems uninitialized (active_stored) values are being used causing huge leaps in duration.

    Let me know if this performance impact seems acceptable. To me, it looks a bit heavier than the t3122 management but not by much.

    Thanks,
    -Michael

• File src/osmo-bsc/bts.c:

  • Patch Set #1, Line 768: if (!lchan->activate.concluded)

    Nevermind this one, I was confused. I think the comment is actually confusing. […]

    Done

  • Patch Set #1, Line 772: timespecsub(&now, &lchan->active_stored, &elapsed);

    what if it's first time? active_stored is {0,0} ? […]

    I believe more checks are needed. In testing on a fully loaded system, I see some massive durations being sent which makes me think we're reaching this point but active_stored hasn't yet been initialized.

• File src/osmo-bsc/bts_trx_vty.c:

  • Patch Set #1, Line 574: vty_out(vty, " Activated %s seconds ago%s",

    "Active for: %d seconds%s" would probably fit more in existing print format.

    Done

To view, visit change 28165. To unsubscribe, or for help writing mail filters, visit settings.