Calculate fraction of bytes spent on answering directory requests
I'm using the following patch to distinguish Directory traffic from OR traffic on my test relay:
diff --git a/src/or/connection.c b/src/or/connection.c index 7b1493b..9eeed63 100644 --- a/src/or/connection.c +++ b/src/or/connection.c @@ -2075,11 +2075,17 @@ connection_buckets_decrement(connection_t *conn, time_t n if (conn->type == CONN_TYPE_EXIT) rep_hist_note_exit_bytes_read(conn->port, num_read); rep_hist_note_bytes_read(num_read, now);
-
log_info(LD_DIR, "Read %d bytes from %s connection " U64_FORMAT ".",
-
num_read, conn_type_to_string(conn->type), -
U64_PRINTF_ARG(conn->global_identifier));} if (num_written > 0) { if (conn->type == CONN_TYPE_EXIT) rep_hist_note_exit_bytes_written(conn->port, num_written); rep_hist_note_bytes_written(num_written, now);
-
log_info(LD_DIR, "Wrote %d bytes to %s connection " U64_FORMAT ".",
-
num_written, conn_type_to_string(conn->type), -
U64_PRINTF_ARG(conn->global_identifier));}
if (connection_counts_as_relayed_traffic(conn, now)) {
As a result, I get a very low fraction of Directory traffic compared to OR traffic. I wonder why that is the case.
In the attachment you can find a histogram of 10-second intervals and the sum of read and written bytes in KiB/s. There are only very few Directory connections with very little bandwidth consumption in that histogram. This is a 500 KiB/s middle node running for 24 hours.