From rusty at rustcorp.com.au  Thu Aug  8 05:56:13 2019
From: rusty at rustcorp.com.au (Rusty Russell)
Date: Thu, 08 Aug 2019 15:26:13 +0930
Subject: [Lightning-dev] Improving Lightning Network Pathfinding Latency
	by Path Splicing and Other Real-Time Strategy Game Techniques
In-Reply-To: <wiX65OpOUF68a4OAqX0LoCqJaMsEAiYvMI9EwSOMXiEZW-wkyzvTpzu1P0GgoC1RKYjLpP-0CM7etvHmkn9tixoNanEFAiGLIjxi3zSgluQ=@protonmail.com>
References: <K5X732PWnUY2qH_s39wSBT1mX_7kt2McTVvFcW1HvpghHVZccNCyZB9R_-HUoAOHX4wqM-T1j7Rs94y8apb3fmlP4AOLM8UwP92VsPYEYxU=@protonmail.com>
	<CAO3Pvs8uR1q-QycLuybh0HYLK6sauueJY8WrsB5jBG+q1LAX5Q@mail.gmail.com>
	<wiX65OpOUF68a4OAqX0LoCqJaMsEAiYvMI9EwSOMXiEZW-wkyzvTpzu1P0GgoC1RKYjLpP-0CM7etvHmkn9tixoNanEFAiGLIjxi3zSgluQ=@protonmail.com>
Message-ID: <87mugkci9u.fsf@rustcorp.com.au>

ZmnSCPxj via Lightning-dev <lightning-dev at lists.linuxfoundation.org> writes:
>> > Typical quotes suggested that commodity hardware would take 2 seconds to
>> > find a route
>> ?
>> Can you provide a reproducible benchmark or further qualify this number (2
>> seconds)?
>
> No reproducible benchmark.

OK, on my digital ocean 2-cpu 4GB ram ntel(R) Xeon(R) CPU E5-2630L 0 @
2.00GHz (which is pretty old hardware), an unoptimized c-lighting
implementation returns random routes on mainnet in:

        Between 3 to 347 msec, mean 220 msec.

That's forking lightning-cli, querying, printing result and exiting
(mainnet, 941 successes, 1353 failures, I ignored the times on
failures since they were usually v fast).

On my Raspberry Pi 2B (compiled with -O3):

        Between 21 to 3330 msec, mean 388 msec.

Cheers,
Rusty.