Optimum mump2p vs GossipSub: Turbocharged Blockchain Data Propagation | Speed Demo — Transcript

Full transcript

1. 0:06In this video we will demonstrate the
2. 0:09relative performance of optimum P2P
3. 0:13against gossip sub. So, gossip sub is a
4. 0:18very well-known wellestablished
5. 0:20peer-to-peer protocol for gossiping and
6. 0:24optimum P2P is a implement is a new al
7. 0:28protocol that does a similar
8. 0:31functionality and which is based on
9. 0:33random linear network codes.
10. 0:36So the main functionality of this
11. 0:39protocol is that you publish some
12. 0:41message on a node and it can be
13. 0:44subscribed by any other other node on
14. 0:47which you receive the message. So in our
15. 0:50setup
16. 0:51we compare these two protocols with 72
17. 0:55nodes in total that are distributed
18. 0:57across the globe. So for gossip sub
19. 1:00protocol we have 36 nodes
20. 1:02and similarly optimum P2P base protocol
21. 1:05is deployed over another 36 nodes. Then
22. 1:09we will look at the size of the message
23. 1:11we publish and also the rate at which we
24. 1:15publish.
25. 1:19For example,
26. 1:21we have here a diagram and geographic
27. 1:24distribution of these nodes. So we have
28. 1:26these nodes for example in optimum P2P
29. 1:29from North America,
30. 1:31Europe
31. 1:33and Asian countries including Australia.
32. 1:35Similarly we have another set of nodes.
33. 1:37So these are almost nearly identical
34. 1:40near distribution of these nodes. So now
35. 1:42we are going to run some test.
36. 1:46Now let's look at the various
37. 1:48parameters. So here on top we show you a
38. 1:52random topic. It's prefixed by custom
39. 1:54topic. size of the message that's being
40. 1:56published it's about 5 5 MB given that
41. 2:00most of the block sizes in different L1
42. 2:03networks are of approximately 2 to 5 MB
43. 2:07large and then we'll be publishing at a
44. 2:09rate of one random messages message per
45. 2:13second and
46. 2:16we are going to see we have two plots
47. 2:18here on the left it will show you that
48. 2:20latency of the optimum P2P base gossip
49. 2:25And on the right we will have the gossip
50. 2:27sub. So now the latencies are in
51. 2:29milliseconds and so we will have average
52. 2:33delay. So now what does it mean by
53. 2:35delay? So a delay of a gossip is that
54. 2:37approximately almost all in this case
55. 2:4095% of the nodes receives the messages
56. 2:42that are published. And in this
57. 2:44experiment we'll publish the same
58. 2:46message to both the networks to some
59. 2:48randomly picked node. So each message
60. 2:52will come originate from a randomly
61. 2:55picked messages from all the nodes from
62. 2:58any of the nodes spread around the node.
63. 3:00So let's start with this.
64. 3:06As we see that so far on the left seven
65. 3:10messages has been published or eight and
66. 3:12then how many are delivered that means
67. 3:13how many have been delivered to at least
68. 3:1695% of the nodes and you can see that
69. 3:19the average latency of the optimum P2P
70. 3:22delivered messages or gossip are much
71. 3:25lower. So which is about close to about
72. 3:27a second on the other hand for gossip
73. 3:29sub it's about 2 seconds. So now also if
74. 3:33you note that how much variability is
75. 3:35there or consistency of delivery you see
76. 3:38that for optimum P2P it's 1/5 or 1/4 of
77. 3:42a second and for gossip sub it's almost
78. 3:46half a second. So next we will change
79. 3:49the size of the message.
80. 3:54increase the message size from 5 MB to 8
81. 3:56MB and also we'll double the rate of
82. 3:58publishing. So that means every second
83. 4:01there will be two messages published
84. 4:02from two randomly picked nodes around
85. 4:04the world. So let's start the
86. 4:06experiment.
87. 4:08And
88. 4:11as you see it's beginning to deliver the
89. 4:14messages. And as you see, optimum P2P
90. 4:18network is delivering messages, you
91. 4:20know, as they are published. Whereas in
92. 4:22gossip sub, we see a lag. Of course, so
93. 4:25far it's getting true, but the lag is
94. 4:27getting bigger. Also the average latency
95. 4:30of optimum P2P has slightly increased to
96. 4:33over by 200 300 milliseconds more
97. 4:37whereas on gossip sub we see it's
98. 4:39continuously increasing and that is
99. 4:41because it's not able to keep up with
100. 4:43the you know delivery of messages with
101. 4:45respect to the rate of publishing. So
102. 4:47essentially you know there is a delay
103. 4:50that's building up. So the more you
104. 4:53publish at a faster rate, more likely
105. 4:55you will have higher and higher delays.
106. 4:58So now we have seen through these
107. 4:59experiments that optimum P2P protocol
108. 5:03performs well under larger messages and
109. 5:05a faster rate of delivery compared to
110. 5:07gossip shop and it's has a much higher
111. 5:10performance compared to gossip shop.