Re: [tcpm] [EXTERNAL] Questions on HyStart++ draft 02

[Yi Huang <[e-mail]>]

I am excited to see Hystart++ was adopted by your QUIC implementation. Thanks Junho!

From: Junho Choi <[e-mail]>
Sent: Monday, April 27, 2020 5:30 PM
To: Yi Huang <[e-mail]>
Cc: Praveen Balasubramanian <[e-mail]>; [e-mail]
Subject: Re: [EXTERNAL] Questions on HyStart++ draft 02

FYI, I made a PR https://github.com/cloudflare/quiche/pull/476<https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fcloudflare%2Fquiche%2Fpull%2F476&data=02%7C01%7Chuanyi%40microsoft.com%7Ce2d04fd13ac043cf8dac08d7eb0b54bd%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637236306220848757&sdata=wTup02uocSEYky5utB2eimcI8zYCmzmfv93SNyGYMX4%3D&reserved=0> for implementing HyStart++ in quiche, based on draft 02.
The algorithm is modified for QUIC a little (e.g. tcp sequence number vs quic packet number) but in general it's same.

Best,

On Tue, Apr 7, 2020 at 5:32 PM Junho Choi <[e-mail]<mailto:[e-mail]>> wrote:
Ok, thanks for clarifying!

On Tue, Apr 7, 2020 at 5:24 PM Yi Huang <[e-mail]<mailto:[e-mail]>> wrote:
Yeah, it does nothing for the first round.

From: Junho Choi <[e-mail]<mailto:[e-mail]>>
Sent: Thursday, April 2, 2020 12:55 AM
To: Yi Huang <[e-mail]<mailto:[e-mail]>>
Cc: Praveen Balasubramanian <[e-mail]<mailto:[e-mail]>>; [e-mail]<mailto:[e-mail]>
Subject: Re: [EXTERNAL] Questions on HyStart++ draft 02

Hi Yi,

Thanks. See inline for my following question.

On Wed, Apr 1, 2020 at 5:04 PM Yi Huang <[e-mail]<mailto:[e-mail]>> wrote:
1. Section 4.2 -- each round is initialized as

   lastRoundMinRTT = currentRoundMinRTT
   currentRoundMinRTT = infinity
   rttSampleCount = 0

In the very beginning of the connection, what is "currentRoundMinRTT" when
there is no previous value?

I am using current rtt (or initial rtt value) and is it ok?

[Yi]: In our implementation, lastRoundMinRTT and currentRoundMinRTT are initialized to 0xffffffff as sentinel values. We only compare lastRoundMinRTT with currentRoundMinRTT when both of them are valid values. Using currRTT seems to essentially behave the same as assigning a sentinel value.

So in the beginning of the connection lastRoundMinRTT is <inf>. When ack is received currentRoundMinRTT = min_rtt but
lastRoundMinRTT is still <inf>. When the 1st round ends, it looks always doing nothing because the following if() is always false?

  if (currentRoundMinRTT >= (lastRoundMinRTT + RttThresh))



2. Section 4.2 -- When used with cubic, CA_cwnd() is based on cubic algorithm I think.
 This means I need to do cubic variables calculation during slow start
 such as K and W_cubic. When is considered as a start of
 congestion avoidance and what W_max will be?

 Currently I use a start of limited slow start as a beginning of congestion recovery
 and use cwnd at the time for W_max. Is my understanding correct?

[Yi]: Yes, that’s exactly what we do in Windows.

Sound great. Thanks.

Best,

--
Junho Choi <junho dot choi at gmail.com<https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgmail.com%2F&data=02%7C01%7Chuanyi%40microsoft.com%7Ce2d04fd13ac043cf8dac08d7eb0b54bd%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637236306220858750&sdata=f8cekJiJx4S625oCkaUDVymXiVPknaarRR9IuLhytfs%3D&reserved=0>>


--
Junho Choi <junho dot choi at gmail.com<https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgmail.com%2F&data=02%7C01%7Chuanyi%40microsoft.com%7Ce2d04fd13ac043cf8dac08d7eb0b54bd%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637236306220868745&sdata=FHoov33SoCU1DWSZZ0n4DqCL0tQVjBrLc2mGBXNxI0Q%3D&reserved=0>>


--
Junho Choi <junho dot choi at gmail.com<https://nam06.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgmail.com%2F&data=02%7C01%7Chuanyi%40microsoft.com%7Ce2d04fd13ac043cf8dac08d7eb0b54bd%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637236306220868745&sdata=FHoov33SoCU1DWSZZ0n4DqCL0tQVjBrLc2mGBXNxI0Q%3D&reserved=0>>