2024-11-25T10:56:38.791Z

<Jose J Palacios-Perez> Update: the following is the 8 OSD, balanced in terms of OSDs (Bill's suggestion), with num reactors from 4 to 5, notice how the performance is better for 4 reactors: with 5 reactors achieves less than 180 kIOPs, whereas with four reactor it achieves slightly more than 180kIOs, with similar latency std dev: https://files.slack.com/files-pri/T1HG3J90S-F081WLV42R5/download/cyan_8osd_bal_osd_4krandread_iops_vs_lat.png

2024-11-25T11:00:26.666Z

<Jose J Palacios-Perez> CPU util is expected to increase as we add more reactors, but notice how the mem utilisation nearly doubles. (I can't find the 8OSD 5 reactor grid to confirm the distribution :cry: ): https://files.slack.com/files-pri/T1HG3J90S-F0828BZ4SMT/download/cyan_8osd_bal_osd_4krandread_osd_cpu.png

2024-11-25T11:00:26.667Z

<Jose J Palacios-Perez> https://files.slack.com/files-pri/T1HG3J90S-F08301E4KCG/download/cyan_8osd_bal_osd_4krandread_osd_mem.png

2024-11-25T11:00:26.671Z

<Jose J Palacios-Perez> https://files.slack.com/files-pri/T1HG3J90S-F082DNKD8SY/download/cyan_8osd_4reactor__bal.png

2024-11-25T16:32:55.733Z

<Jose J Palacios-Perez> Out of curiosity, could you please point me to the code where the memory allocation of the Seastar reactors take place? I was guessing seastar/src/core/memory.cc