Analyzing the Costs and Benefits of DNS, DoT, and DoH for the … · 2019. 10. 30. · Analyzing...
Transcript of Analyzing the Costs and Benefits of DNS, DoT, and DoH for the … · 2019. 10. 30. · Analyzing...
-
Analyzing the Costs and Benefits of DNS, DoT, and DoH for the Modern Web
Austin Hounsel* Kevin Borgolte* Paul Schmitt*Jordan Holland* Nick Feamster†
Princeton University* University of Chicago†1
-
DNS Privacy Has Become a Significant Concern
● On-path observers can spy on traditional DNS (Do53)
● Two protocols have been proposed to encrypt DNS traffico DNS-over-TLS (DoT)
o DNS-over-HTTPS (DoH)
2
-
Contributions
● Extensive performance study of Do53, DoT, and DoHo Query response times
o Page load times
o Emulated network conditions
● Measurements from five global vantage points
3
-
Unexpected Finding
● Despite higher response times, page load times with
encrypted DNS transports can be faster than Do53
4
-
DNS Responses from Cloudflare at Ohio
5
0 100 200D1S 5HsSRnsH 7iPH (Ps)
0.00
0.25
0.50
0.75
1.00
3rRb
Dbili
ty
ClRudflDrH DR53ClRudflDrH DRH
ClRudflDrH DR7DHfDult DR53
0 15000
1
DoH catches up to Do53 in tailDoH outperforms
25% of DoT queries
-
DNS Responses from Google at Ohio
6
0 100 200D1S 5HsSRnsH 7iPH (Ps)
0.00
0.25
0.50
0.75
1.00
3rRb
Dbili
ty
DHfDult DR53GRRglH DR53
GRRglH DRHGRRglH DR7
0 15000
1DoH catches up
to Do53/DoT
-
DNS Responses from Quad9 at Ohio
7
0 100 200D1S 5HsSRQsH 7iPH (Ps)
0.00
0.25
0.50
0.75
1.00
3rRb
Dbili
ty
DHfDult DR53QuDd9 DR53
QuDd9 DRHQuDd9 DR7
0 15000
1
DoH outperforms DoT for almost
all queries
DoH catches up to Do53 in the
tail
-
Takeaway: DoH Can Outperform Do53
● DoH has a higher mean but lower variance
● Several possible explanationso HTTP caching at the edge
o Wire format caching
8
-
Emulated Cellular Conditions
● We emulated 4G, lossy 4G, and 3G network conditionso DoH and DoT are starting to be offered on phones
o Performance may be significantly different
9
-
Page Loads with Cloudflare at Ohio
10
Diff ≥ 1s 0.1s ≤ Diff < 1s 0.03s ≤ Diff < 0.1s -0.03s < Diff < 0.03s -0.1s < Diff ≤ -0.03s -1s < Diff ≤ -0.1s Diff ≤ -1s
-10 -1 0 1 103DgH LoDd TLPH DLffHrHncH (sHconds)
0.0
0.2
0.4
0.6
0.8
1.0
3rob
DbLlL
ty
CloudflDrH DoH - CloudflDrH Do53
-10 -1 0 1 103Dge LoDd TLPe DLfference (seconds)
0.0
0.2
0.4
0.6
0.8
1.0
3rob
DbLlL
ty
CloudflDre DoT - CloudflDre Do53
DoT was 1ms slower than Do53
DoH was 19ms slower than Do53
-
Page Loads with Cloudflare at Ohio (4G)
11
Diff ≥ 1s 0.1s ≤ Diff < 1s 0.03s ≤ Diff < 0.1s -0.03s < Diff < 0.03s -0.1s < Diff ≤ -0.03s -1s < Diff ≤ -0.1s Diff ≤ -1s
-10 -1 0 1 103DgH LoDd TLPH DLffHrHncH (sHconds)
0.00.20.40.60.81.0
3rob
DbLlL
ty
CloudflDrH DoH - CloudflDrH Do53
-10 -1 0 1 103Dge LoDd TLPe DLfference (seconds)
0.0
0.2
0.4
0.6
0.8
1.0
3rob
DbLlL
ty
CloudflDre DoT - CloudflDre Do53
DoT was 1ms faster than Do53
DoH was 70ms slower than Do53
-
Page Loads with Cloudflare at Ohio (Lossy 4G)
12
Diff ≥ 1s 0.1s ≤ Diff < 1s 0.03s ≤ Diff < 0.1s -0.03s < Diff < 0.03s -0.1s < Diff ≤ -0.03s -1s < Diff ≤ -0.1s Diff ≤ -1s
-10 -1 0 1 103DgH LoDd TLPH DLffHrHncH (sHconds)
0.0
0.2
0.4
0.6
0.8
1.0
3rob
DbLlL
ty
CloudflDrH DoH - CloudflDrH Do53
-10 -1 0 1 103Dge LoDd TLPe DLfference (seconds)
0.0
0.2
0.4
0.6
0.8
1.0
3rob
DbLlL
ty
CloudflDre DoT - CloudflDre Do53
DoT was 62ms faster than Do53
DoH was 20ms faster than Do53
-
Page Loads with Cloudflare at Ohio (3G)
13
Diff ≥ 1s 0.1s ≤ Diff < 1s 0.03s ≤ Diff < 0.1s -0.03s < Diff < 0.03s -0.1s < Diff ≤ -0.03s -1s < Diff ≤ -0.1s Diff ≤ -1s
-10 -1 0 1 103DgH LoDd TLPH DLffHrHncH (sHconds)
0.00.20.40.60.81.0
3rob
DbLlL
ty
CloudflDrH DoH - CloudflDrH Do53
-10 -1 0 1 103Dge LoDd TLPe DLfference (seconds)
0.00.20.40.60.81.0
3rob
DbLlL
ty
CloudflDre DoT - CloudflDre Do53
DoT was 197ms slower than Do53
DoH was 260ms slower than Do53
-
Takeaway: TCP Helps Page Load Times
● TCP packets can be retransmitted after 2x RTT
● Timeout of Do53 is set to 5 seconds by default in Linux
14
-
Summary
● Extensive performance study of Do53, DoT, and DoHo Query response times
o Page load times
o Emulated network conditions
● Future work: performance analyses over diverse networks
15