Weird performance on the load

[gobwas]

Hello!

Im currently researching technologies for some web app, and have wrote a simple http server as you show in your example:

extern crate hyper;

use std::io::Write;

use hyper::Server;
use hyper::server::Request;
use hyper::server::Response;
use hyper::net::Fresh;

fn hello(_: Request, res: Response<Fresh>) {
    let mut res = res.start().unwrap();
    res.write_all(b"Hello World!").unwrap();
    res.end().unwrap();
}

fn main() {
    Server::http("127.0.0.1:4000").unwrap().handle(hello);
}

Then I used JMeter for emulate some load, and thats what I've found, for example, in comparsion with Node.js:

Samples	Average	Median	90%	95%	99%	Min	Max	Throughput /sec
hyper.rs	2500	595	0	4	5619	10859	0	12387
node.js	2500	2	2	6	7	9	0	30

50'th and 90'th percentiles are very fast, and faster than node, but, then, 95 and higher - is much lower.
It means, that 95% of requests are faster than node, but 5% much slower.

Then I've looked at detailed results of requests, and that is what I have found (more attention on the Duration column - it repeat high latency in period of ~50 requests):

Sample	Thread	Duration(ms)	Bytes	Latency
1	Pure hello 1-2	4	106	4
2	Pure hello 1-1	11	106	11
3	Pure hello 1-3	2	106	2
4	Pure hello 1-4	4	106	4
5	Pure hello 1-5	6	106	6
6	Pure hello 1-1	0	106	0
7	Pure hello 1-2	1	106	1
8	Pure hello 1-3	1	106	1
9	Pure hello 1-4	1	106	1
10	Pure hello 1-5	1	106	1
11	Pure hello 1-1	0	106	0
12	Pure hello 1-2	0	106	0
13	Pure hello 1-3	1	106	1
14	Pure hello 1-4	1	106	1
15	Pure hello 1-5	1	106	1
16	Pure hello 1-2	0	106	0
17	Pure hello 1-1	0	106	0
18	Pure hello 1-3	0	106	0
19	Pure hello 1-4	0	106	0
20	Pure hello 1-5	0	106	0
21	Pure hello 1-2	0	106	0
22	Pure hello 1-1	1	106	1
23	Pure hello 1-3	1	106	1
24	Pure hello 1-4	1	106	1
25	Pure hello 1-5	1	106	1
26	Pure hello 1-2	1	106	1
27	Pure hello 1-1	1	106	1
28	Pure hello 1-3	0	106	0
29	Pure hello 1-4	0	106	0
30	Pure hello 1-5	0	106	0
31	Pure hello 1-2	0	106	0
32	Pure hello 1-1	0	106	0
33	Pure hello 1-4	0	106	0
34	Pure hello 1-3	0	106	0
35	Pure hello 1-5	0	106	0
36	Pure hello 1-2	0	106	0
37	Pure hello 1-1	0	106	0
38	Pure hello 1-4	0	106	0
39	Pure hello 1-3	0	106	0
40	Pure hello 1-5	0	106	0
41	Pure hello 1-2	1	106	1
42	Pure hello 1-1	0	106	0
43	Pure hello 1-4	0	106	0
44	Pure hello 1-3	0	106	0
45	Pure hello 1-5	0	106	0
46	Pure hello 1-2	0	106	0
47	Pure hello 1-1	0	106	0
48	Pure hello 1-4	0	106	0
49	Pure hello 1-3	0	106	0
50	Pure hello 1-5	0	106	0
51	Pure hello 1-6	448	106	448
52	Pure hello 1-7	480	106	480
53	Pure hello 1-8	478	106	478
54	Pure hello 1-9	476	106	476
55	Pure hello 1-10	473	106	473
56	Pure hello 1-6	0	106	0
57	Pure hello 1-8	0	106	0
58	Pure hello 1-7	1	106	1
59	Pure hello 1-10	0	106	0
60	Pure hello 1-9	0	106	0
61	Pure hello 1-6	0	106	0
62	Pure hello 1-7	1	106	1
63	Pure hello 1-8	1	106	1
64	Pure hello 1-9	0	106	0
65	Pure hello 1-10	0	106	0
66	Pure hello 1-6	1	106	1
67	Pure hello 1-7	0	106	0
68	Pure hello 1-8	0	106	0
69	Pure hello 1-9	1	106	1
70	Pure hello 1-10	1	106	1
71	Pure hello 1-6	0	106	0
72	Pure hello 1-7	0	106	0
73	Pure hello 1-8	1	106	1
74	Pure hello 1-9	1	106	1
75	Pure hello 1-10	0	106	0
76	Pure hello 1-6	1	106	1
77	Pure hello 1-8	0	106	0
78	Pure hello 1-7	0	106	0
79	Pure hello 1-10	0	106	0
80	Pure hello 1-9	0	106	0
81	Pure hello 1-6	1	106	1
82	Pure hello 1-8	1	106	1
83	Pure hello 1-7	1	106	1
84	Pure hello 1-10	1	106	1
85	Pure hello 1-9	1	106	1
86	Pure hello 1-8	1	106	1
87	Pure hello 1-6	2	106	2
88	Pure hello 1-7	2	106	2
89	Pure hello 1-9	7	106	7
90	Pure hello 1-10	8	106	8
91	Pure hello 1-8	0	106	0
92	Pure hello 1-6	0	106	0
93	Pure hello 1-7	0	106	0
94	Pure hello 1-9	0	106	0
95	Pure hello 1-10	0	106	0
96	Pure hello 1-8	0	106	0
97	Pure hello 1-6	0	106	0
98	Pure hello 1-7	0	106	0
99	Pure hello 1-10	0	106	0
100	Pure hello 1-9	0	106	0
101	Pure hello 1-13	895	106	895
102	Pure hello 1-12	898	106	898
103	Pure hello 1-11	903	106	903
104	Pure hello 1-14	893	106	893
105	Pure hello 1-15	888	106	888
106	Pure hello 1-196	116	1791	0
107	Pure hello 1-12	0	106	0
108	Pure hello 1-13	0	106	0
109	Pure hello 1-11	0	106	0
110	Pure hello 1-14	0	106	0
111	Pure hello 1-15	1	106	1
112	Pure hello 1-12	0	106	0
113	Pure hello 1-11	1	106	1
114	Pure hello 1-13	0	106	0
115	Pure hello 1-15	0	106	0
116	Pure hello 1-14	0	106	0
117	Pure hello 1-11	1	106	1
118	Pure hello 1-12	1	106	1
119	Pure hello 1-13	1	106	1
120	Pure hello 1-14	0	106	0
121	Pure hello 1-15	0	106	0
122	Pure hello 1-12	1	106	1
123	Pure hello 1-13	0	106	0
124	Pure hello 1-11	1	106	1
125	Pure hello 1-15	1	106	1
126	Pure hello 1-14	1	106	1
127	Pure hello 1-13	1	106	1
128	Pure hello 1-12	1	106	1
129	Pure hello 1-11	0	106	0
130	Pure hello 1-15	1	106	1
131	Pure hello 1-14	1	106	1
132	Pure hello 1-12	0	106	0
133	Pure hello 1-13	0	106	0
134	Pure hello 1-11	0	106	0
135	Pure hello 1-14	0	106	0
136	Pure hello 1-15	0	106	0
137	Pure hello 1-13	0	106	0
138	Pure hello 1-12	0	106	0
139	Pure hello 1-11	0	106	0
140	Pure hello 1-15	1	106	1
141	Pure hello 1-14	1	106	1
142	Pure hello 1-13	1	106	1
143	Pure hello 1-12	0	106	0
144	Pure hello 1-11	0	106	0
145	Pure hello 1-15	0	106	0
146	Pure hello 1-14	0	106	0
147	Pure hello 1-13	1	106	1
148	Pure hello 1-11	1	106	1
149	Pure hello 1-12	1	106	1
150	Pure hello 1-15	1	106	1
151	Pure hello 1-14	1	106	1
152	Pure hello 1-16	1117	106	1117
153	Pure hello 1-18	1110	106	1110
154	Pure hello 1-17	1114	106	1114
155	Pure hello 1-20	1132	106	1132
156	Pure hello 1-19	1139	106	1139
157	Pure hello 1-18	0	106	0
158	Pure hello 1-16	0	106	0
159	Pure hello 1-17	0	106	0
160	Pure hello 1-20	0	106	0
161	Pure hello 1-19	0	106	0
162	Pure hello 1-18	0	106	0
163	Pure hello 1-16	1	106	1
164	Pure hello 1-17	1	106	1
165	Pure hello 1-19	1	106	1
166	Pure hello 1-20	1	106	1
167	Pure hello 1-16	1	106	1
168	Pure hello 1-17	1	106	1

... and so on.

Is it a bug or feature? 😄

Thanks )

[gobwas]

Also, FYI, the "Thread group" of JMeter is configured as:

[seanmonstar]

I assume it's related to 1) no timeouts #368, and 2) the use of synchronous IO, which pretty much just makes No. 1 hurt more.

[gobwas]

How and when could it be solved? I see that #368 is closed..

[seanmonstar]

No, it's still open. Just several issues referencing it are closed.

For timeouts, see #315. For async IO, see #395.

[gobwas]

Oops, sorry 👍
Thank you!