Yingzhen-ietf / BGP-QUIC

3 stars 1 forks source link

Will BGP over QUIC ever run out of bytes? #15

Closed Yingzhen-ietf closed 9 months ago

Yingzhen-ietf commented 1 year ago

Streams and connections in QUIC can each carry a maximum of 2^62-1 bytes in each direction due to encoding limitations on stream offsets and connection flow control limits. In case this limit is reached, a new connection needs to be established. BoQ is a long lasting connection, will it run out of bytes count?

jhaas-pfrc commented 9 months ago

2^62 - 1 is roughly 4.6 * 10^18 Let's pick an example of something as busy as the RIPE RIS servers. Looking at https://data.ris.ripe.net/rrc00/2023.12/ which covers the month of December 2023, the updates files are on average less than 10M in zipped form. BGP compresses very well, so if we presume the 4-5M expands to 2x to 10M that can help us with a heuristic.

There are 8928 5 minute intervals in a 31 day month.

At 10MB for all 5 minute samples in that month, this gives us 8.9*10^10.

Presuming multiplicative expansion for a year (times 12), we're still significantly under 10^18.

I think we're good.

Even if we're not good, closing a stream and opening a new one would bring us back up. If BGP over QUIC sessions are long-lived enough to require coordination of this, we're far more successful than I'd expect.

Yingzhen-ietf commented 9 months ago

LOL. I totally agree with you. We'd be happy to see this problem.

Yingzhen

On Tue, Jan 16, 2024 at 11:56 AM jhaas-pfrc @.***> wrote:

2^62 - 1 is roughly 4.6 * 10^18 Let's pick an example of something as busy as the RIPE RIS servers. Looking at https://data.ris.ripe.net/rrc00/2023.12/ which covers the month of December 2023, the updates files are on average less than 10M in zipped form. BGP compresses very well, so if we presume the 4-5M expands to 2x to 10M that can help us with a heuristic.

There are 8928 5 minute intervals in a 31 day month.

At 10MB for all 5 minute samples in that month, this gives us 8.9*10^10.

Presuming multiplicative expansion for a year (times 12), we're still significantly under 10^18.

I think we're good.

Even if we're not good, closing a stream and opening a new one would bring us back up. If BGP over QUIC sessions are long-lived enough to require coordination of this, we're far more successful than I'd expect.

— Reply to this email directly, view it on GitHub https://github.com/Yingzhen-ietf/BGP-QUIC/issues/15#issuecomment-1894417150, or unsubscribe https://github.com/notifications/unsubscribe-auth/AURDC42N75IGJEZXHSMRJNLYO3LNJAVCNFSM6AAAAAAVNVQRXGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQOJUGQYTOMJVGA . You are receiving this because you authored the thread.Message ID: @.***>