Xmas combo

[benaadams]

Best yet... (though will probably have to break it down into pieces?)

Azure G4 Peaking 2.7 GBps send 1.1GBps receive

344,334 rps Non pipelined w/ 3.44ms latency (32 threads and 1024 connections)
2,398,936 rps 16 pipelined w/ 5.66ms latency (32 threads and 1024 connections)

(3.9 - 4.2 Gbps send / 1.4 - 1.5 Gbps receive when pipelined 512)

[benaadams]

Running 30s test @ http://.../plaintext
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 6.50ms 16.64ms 288.14ms 96.96%
Req/Sec 75.44k 7.43k 183.60k 86.71%
69929367 requests in 30.08s, 8.60GB read
Socket errors: connect 35, read 0, write 0, timeout 0
Requests/sec: 2324948.53
Transfer/sec: 292.68MB

[benaadams]

Pipelined 16

Running 30s test @ http://..../plaintext
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 5.66ms 13.27ms 366.96ms 97.21%
Req/Sec 77.84k 7.30k 172.61k 82.76%
72158246 requests in 30.08s, 8.87GB read
Socket errors: connect 35, read 0, write 0, timeout 0
Requests/sec: 2398936.64
Transfer/sec: 301.99MB

Non-pipelined

Running 30s test @ http://.../plaintext
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 3.44ms 6.30ms 288.05ms 97.78%
Req/Sec 10.84k 1.59k 55.05k 86.35%
10364378 requests in 30.10s, 1.27GB read
Requests/sec: 344334.25
Transfer/sec: 43.35MB

[cesarblum]

This is a lot in a single PR. Is it all really part of the same change or could you break this up into multiple PRs?

[benaadams]

I think its more of a journey that a PR :)

I should be able to split it in to separate PRs though have some of them build of a merge of another?

e.g. Only pooled blocks -> Expose pointer -> Removed fixed (in lots of places)

[benaadams]

After last two

Running 30s test @ http://.../plaintext
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 5.63ms 12.10ms 406.38ms 96.73%
Req/Sec 77.37k 7.53k 258.40k 86.85%
74024392 requests in 30.10s, 9.10GB read
Requests/sec: 2459192.17
Transfer/sec: 309.58MB

[benaadams]

Not sure about last commit, but...

Running 30s test @ http://..../plaintext
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 6.68ms 17.92ms 353.24ms 96.80%
Req/Sec 78.43k 7.62k 146.88k 83.62%
75113786 requests in 30.10s, 9.23GB read
Requests/sec: 2495536.14
Transfer/sec: 314.15MB

[benaadams]

256 latency is crazy

Running 30s test @ http://.../plaintext
32 threads and 256 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 1.49ms 5.73ms 211.36ms 88.81%
Req/Sec 78.07k 6.52k 151.92k 80.45%
74725034 requests in 30.10s, 9.19GB read
Requests/sec: 2482757.39
Transfer/sec: 312.54MB

[benaadams]

[benaadams]

Running 1m test @ http://.../plaintext
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 5.09ms 12.34ms 310.39ms 97.63%
Req/Sec 86.81k 7.82k 170.18k 79.29%
160832007 requests in 1.00m, 19.77GB read
Socket errors: connect 35, read 0, write 0, timeout 0
Requests/sec: 2676090.47
Transfer/sec: 336.88MB

with dotnet/extensions#74 and aspnet/HttpAbstractions#521

[benaadams]

Running 1m test @ http://.../plaintext
32 threads and 256 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 1.20ms 4.92ms 228.28ms 91.26%
Req/Sec 87.25k 6.86k 144.16k 76.85%
166848206 requests in 1.00m, 20.51GB read
Requests/sec: 2776712.83
Transfer/sec: 349.55MB

[benaadams]

Running 1m test @ http://.../plaintext
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 4.82ms 12.41ms 281.62ms 97.75%
Req/Sec 89.63k 16.84k 179.39k 91.52%
171131325 requests in 1.00m, 21.04GB read
Socket errors: connect 35, read 0, write 0, timeout 0
Requests/sec: 2848191.98
Transfer/sec: 358.54MB

Running 1m test @ http://.../plaintext
32 threads and 256 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 1.12ms 3.71ms 209.34ms 89.65%
Req/Sec 90.31k 6.54k 135.20k 76.12%
172694464 requests in 1.00m, 21.23GB read
Requests/sec: 2874816.92
Transfer/sec: 361.90MB

[halter73]

Is there any particular reason that the ValueTask inclusion didn't make the combo?

[benaadams]

@halter73 not really :)

There is a change to OutputSocket in this that is a little dodgy; sometimes good for throughput sometimes really bad. Have to rework it - but will do that as an individual PR

[halter73]

@benaadams Since you've split this into smaller PRs, are we OK to close this now?

[benaadams]

Yeah; though is a bunch more it this that I haven't done PRs for; but I'll remember them.