[WIP] debug_traceBlockByNumber way too slow

Tom M — Tue, 12 Dec 2023 10:18:30 +0000

hey, running xdc (full syncmode, gcmode archive, not mining) on multiple nodes. We're experiencing weird issue with debug_traceBlockByNumber calls.

on 24core AMD (AMD EPYC 7413 24-Core Processor) it takes almost 40secs to get the response, on 12core (Intel(R) Xeon(R) E-2386G CPU @ 3.50GHz) the same query takes 15secs. All the same version, data and configuration, we're using nvme disks.

query:

{
    "jsonrpc":"2.0",
        "method":"debug_traceBlockByNumber",
        "params":["0x41A0126", {"tracer": "callTracer", "timeout":"60s"}],
        "id":1
}

24 cores:

     time_namelookup:  0.000013s
        time_connect:  0.000064s
     time_appconnect:  0.000000s
    time_pretransfer:  0.000084s
       time_redirect:  0.000000s
  time_starttransfer:  33.867162s
                     ----------
          time_total:  33.867536s

12cores:

     time_namelookup:  0.000038s
        time_connect:  0.000198s
     time_appconnect:  0.000000s
    time_pretransfer:  0.000260s
       time_redirect:  0.000000s
  time_starttransfer:  15.226504s
                     ----------
          time_total:  15.226841s

we have issue even with these 15secs, is that a common response time?

Any idea about possible tuning? Saw that cache is disabled, no LRU cache, no new tracers as in geth

thanks

Developers Forum for XinFin XDC Network: Tom M

[WIP] debug_traceBlockByNumber way too slow