Things you notice looking at tcpdumps: DynamoDB sends a TCP ACK for the segment containing a request, then sends its response a few ms later. It sure would be nice if there were a configuration option which allowed them to save bandwidth (+CPU time) by delaying that ACK.
-
-
Replying to @cperciva
I'm not sure about the trade-off; letting the client kernel free up the buffered segments while waiting might still be better over-all in nearly all cases. But either way, a TCP_QUICKACK contribution to Amazon s2n would be welcome :-)
3 replies 0 retweets 1 like -
Replying to @colmmacc
In this case, sending an ACK immediately allows clients to free up ~ 1 kB of TCP buffers ~ 10ms sooner. I can't imagine any system where that's valuable enough to justify the bandwidth and CPU cost of sending that extra ACK packet.
2 replies 0 retweets 0 likes -
Also... these connections were HTTP, so s2n wasn't involved anyway. ;-)
1 reply 0 retweets 0 likes -
Replying to @cperciva
*blink* Who are you and what have you done with the real Colin Percival? You're *not* using encryption?
1 reply 0 retweets 1 like -
Replying to @colmmacc
Who said I'm not using encryption? ;-) There are many situations where I prefer to use end-to-end encryption and HTTP rather than using HTTPS and exposing myself to TLS stack bugs.
1 reply 0 retweets 0 likes
Well you know better than anyone what's being exposed even when the objects in the tables are themselves are encrypted. Still, it makes me shudder a bit. Might make defending against more granular Identification and fingerprinting attacks, if they are in the threat model.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.