Add period:100 reporting

content/articles/2025-08-10-ctlog-2.md

@@ -542,10 +542,12 @@ sunlight_sqlite_update_duration_seconds{quantile="0.99"} 0.016859223
 
 Comparing the throughput at 4'400/s with that first test of 100/s, I expect and can confirm a
 significant increase in all of these metrics. The 99th percentile addchain is now 1889ms (up from
-207ms) and the sequencing duration is now 1111ms (up from 109ms).  I fiddle a little bit with
-Sunlight's configuration file, notably the `period` and `poolsize`.  This does not seem to matter
-much. Setting for example `period:2000` and `poolsize:15000` still yields pretty much the same
-throughput:
+207ms) and the sequencing duration is now 1111ms (up from 109ms). 
+
+#### Sunlight: Effect of period
+
+I fiddle a little bit with Sunlight's configuration file, notably the `period` and `poolsize`.
+First I set `period:2000` and `poolsize:15000`, which yields pretty much the same throughput:
 
 ```
 pim@ctlog-test:/etc/sunlight$ while :; do curl -ksS https://ctlog-test.lab.ipng.ch:1443/checkpoint | grep -E '^[0-9]+$'; sleep 60; done
@@ -572,6 +574,34 @@ sunlight_sequencing_duration_seconds{log="sunlight-test",quantile="0.99"} 1.5968
 sunlight_sqlite_update_duration_seconds{quantile="0.99"} 0.010847308
 ```
 
+Then I also set a `period:100` and `poolsize:15000`, which does improve a bit:
+
+```
+pim@ctlog-test:/etc/sunlight$ while :; do curl -ksS https://ctlog-test.lab.ipng.ch:1443/checkpoint | grep -E '^[0-9]+$'; sleep 60; done
+560654
+950524
+1324645
+1720362
+```
+
+With the same generated load of 10'000/sec with a 10% duplication rate, I am still offering roughly
+9'000/sec of unique certificates, and I'm seeing `(1720362 - 560654)/180` or about 6'440/sec come
+through, which is a fair bit better, at the expense of more disk activity. At this rate and with
+`period:100`, the latency tail looks like this:
+
+```
+pim@ctlog-test:/etc/sunlight$ curl -ksS https://ctlog-test.lab.ipng.ch/metrics | egrep 'seconds.*quantile=\"0.99\"'
+sunlight_addchain_wait_seconds{log="sunlight-test",quantile="0.99"} 1.616046445
+sunlight_cache_get_duration_seconds{log="sunlight-test",quantile="0.99"} 7.5123e-05
+sunlight_cache_put_duration_seconds{log="sunlight-test",quantile="0.99"} 0.534935803
+sunlight_fs_op_duration_seconds{log="sunlight-test",method="discard",quantile="0.99"} 0.000377273
+sunlight_fs_op_duration_seconds{log="sunlight-test",method="fetch",quantile="0.99"} 4.8893e-05
+sunlight_fs_op_duration_seconds{log="sunlight-test",method="upload",quantile="0.99"} 0.054685991
+sunlight_http_request_duration_seconds{endpoint="add-chain",log="sunlight-test",quantile="0.99"} 1.946445877
+sunlight_sequencing_duration_seconds{log="sunlight-test",quantile="0.99"} 0.980602185
+sunlight_sqlite_update_duration_seconds{quantile="0.99"} 0.018385831
+```
+
 ***Conclusion***: Sunlight on POSIX can reliably handle 4'400/s (with a duplicate rate of 10%) on
 this setup.
 
@@ -589,9 +619,9 @@ current write-load (which is about 250/s).
      solutions, taking into account that TesseraCT runs MariaDB (which my setup did not use ZFS
      for), while Sunlight uses sqlite3 on the ZFS pool.
 
-***Notable***: Sunlight POSIX and S3 performance is roughly identical (both handle about 5'000/sec),
-while TesseraCT POSIX performance (12'000/s) is significantly better than its S3 (800/s).  Some
-other observations:
+***Notable***: Sunlight POSIX and S3 performance is roughly identical (both handle about
+5'000/sec), while TesseraCT POSIX performance (12'000/s) is significantly better than its S3
+(800/s). Some other observations:
 
 *    Sunlight has a very opinionated configuration, and can run multiple logs with one configuration
      file and one binary. Its configuration was a bit constraining though, as I could not manage to