Use a single lock for writing to ConcurrentPipeWriter

[jkotalik]

Follow up to #12081

Before:

All results:
| Description |       RPS | CPU (%) | Memory (MB) | Avg. Latency (ms) | Startup (ms) | First Request (ms) | Latency (ms) | Errors |
| ----------- | --------- | ------- | ----------- | ----------------- | ------------ | ------------------ | ------------ | ------ |
|           √ | 4,063,529 |     100 |         106 |               0.9 |       230.23 |            46.8208 |       0.1244 |      0 |
|           √ | 4,092,813 |      99 |         109 |              1.26 |     230.6499 |            46.6005 |       0.1598 |      0 |
|           √ | 4,036,415 |      99 |         103 |              1.03 |      229.846 |            46.5652 |       0.1287 |      0 |

RequestsPerSecond:           4,064,252
Max CPU (%):                 99
WorkingSet (MB):             106
Avg. Latency (ms):           1.06
Startup (ms):                230
First Request (ms):          46.66
Latency (ms):                0.14
Total Requests:              61,368,471
Duration: (ms)               15,100
Socket Errors:               0
Bad Responses:               0
SDK:                         3.0.100-preview8-013304
Runtime:                     3.0.0-preview8-27914-06
ASP.NET Core:                3.0.0-preview8.19367.2

After:

All results:
| Description |       RPS | CPU (%) | Memory (MB) | Avg. Latency (ms) | Startup (ms) | First Request (ms) | Latency (ms) | Errors |
| ----------- | --------- | ------- | ----------- | ----------------- | ------------ | ------------------ | ------------ | ------ |
|           √ | 4,106,103 |      99 |         102 |              1.09 |     243.4135 |            85.7877 |       0.1238 |      0 |
|           √ | 4,122,374 |      99 |         110 |              1.13 |     244.1361 |            86.1066 |       0.1254 |      0 |
|           √ | 4,155,889 |      99 |         106 |              1.14 |     242.9591 |            86.0695 |       0.1264 |      0 |

RequestsPerSecond:           4,128,122
Max CPU (%):                 99
WorkingSet (MB):             106
Avg. Latency (ms):           1.12
Startup (ms):                244
First Request (ms):          85.99
Latency (ms):                0.13
Total Requests:              62,333,715
Duration: (ms)               15,100
Socket Errors:               0
Bad Responses:               0
SDK:                         3.0.100-preview8-013305
Runtime:                     3.0.0-preview8-27914-06
ASP.NET Core:                3.0.0-preview8.19367.2

The first request numbers are weird. Both #12265 and this change show a 2x regression on the first request. I feel like there is something that slows down first request when passing in output files.

I wasn't sure if I could remove the lock in Http2Framewriter. I'll follow up with @halter73 with regards to that.

[halter73]

👍

[halter73]

I'd appreciate it ff we could wait for the ~2% regression this fixes to show up on aka.ms/aspnet/benchmarks before merging.

I feel like there is something that slows down first request when passing in output files.

That's why when I run my comparison, I use output files for both master and the branch I'm testing.

[halter73]

Also, are you running with the physical windows server and linux load client? I was getting much closer to 5 million RPS when I benchmarked lock sharing, but my baseline was also much higher.

[jkotalik]

Linux for both.