2024-11-07T10:21:10.938Z

<Jose J Palacios-Perez> The following low tech format is easier to compare configurations, for example the following is for 3 OSD, 10 reactors per OSD, using only "physical" CPU cores (HT disabled), which corresponds to fourth row of the second table above. Compared with the similar case when HT is enabled (rhs):: https://files.slack.com/files-pri/T1HG3J90S-F07V2L1PAG6/download/crimson_3osd_10reactor_disable_ht.png

2024-11-07T10:21:10.939Z

<Jose J Palacios-Perez> https://files.slack.com/files-pri/T1HG3J90S-F07V6D8KEAZ/download/crimson_3osd_10reactor_enable_ht.png

2024-11-07T10:23:31.800Z

<Jose J Palacios-Perez> I'll extend the tool to deal with `taskset` range output since the field PSR from `ps` only shows the latest CPU id

2024-11-07T10:26:12.919Z

<Jose J Palacios-Perez> sorry, forgot to mention: Red 'R' stands for Reactor thread (including "syscal", "crimson" and a single "log"), green 'A' for Alien worker threads, and blue 'B' for Bluestore threads

2024-11-07T10:31:02.203Z

<Jose J Palacios-Perez> I've normally been using CPU ids 48-55 for FIO (exclusively via `taskset`), so in principle they might be contending with some Alien+BS threads

2024-11-07T10:33:30.107Z

<Jose J Palacios-Perez> The following is an example of a configuration that we *could* test, but not really sure will be worth pursuing: is too unbalanced: https://files.slack.com/files-pri/T1HG3J90S-F07V9AHKLLA/download/crimson_3osd_32reactor_enable_ht.png

2024-11-07T12:35:16.034Z

<cfanz> Hi, guys. Could any one give me some clue about how inline-block stored in segment-journal are trimmed? It seems neither `trim_dirty()` nor `trim_alloc()` will deal with those extents. Thanks in advance!