trafficserver-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Susan Hinrichs <>
Subject Proposed change in default cipher_suite list for ATS 6.0
Date Thu, 18 Jun 2015 21:49:25 GMT
We are planning on changing the default cipher_suite list as we move to 
ATS 6.0.  The jira outlines the discussion on this issue

Here is the last entry of the jira with the proposal and rationale.

Ran some tests on a production box in Y!  Based on those results, I 
suggest the following cipher string.


The upshot is that we remove RC4, add new ciphers, and rearrange the 
list to give preference to cipher attributes in the following order: 
PFS, then GCM, then stronger SHA, then stronger AES.  3DES is at the end 
to scoop up the remainders.

We tested in the Y! environment which tends to have a wide variety of 
clients.  Removing RC4 did not seem to significantly impact handshake 
success rate.  CBC algorithms are also concerning, but if we care about 
out-of-the-box experience it looks like the CBC algorithms need to stick 
around for a while longer.

Here are details of the test

With Y! original cipher string
0.0102% ssl_error_ssl

The number of DES-CBC3-SHA sessions was negligible (45).  The Y! initial 
configuration has one RC4 algorithm listed kind of early, so the RC4 
percentage was around 30% as [~davet] noted in an earlier comment.

With proposed default cipher string running for an hour
0.009% ssl_error_ssl

The percentage of DES-CBC3-SHA sessions grew to 0.9% of sessions. In my 
experiment, it was impossible to isolate the CPU impact of this change.  
To test a new cipher without updating all the machines in the production 
pod, I removed the test box from the SSL session sharing communication.  
The test box experienced around a 30% increase in CPU utilization, but I 
think that can be mostly attributed to increased session negotiation 
since it did not know about the sessions negotiated by other machines in 
the pod.

We did one experiment with the RC4 ciphers added after DES-CBC3 as 
another measure of how many clients are only willing to do RC4. After 
about an hour, 2 RC4 sessions were started.

510932 = Total Successful Handshakes

Percentage of various cipher's negotiated

# Start with PFS/GCM ciphers.  Give slight preference to AES256 over 
AES128, and prefer stronger SHA
0%      ECDHE-ECDSA-AES256-GCM-SHA384:
4.2%   ECDHE-RSA-AES256-GCM-SHA384:
0%      ECDHE-ECDSA-AES128-GCM-SHA256:
30.6% ECDHE-RSA-AES128-GCM-SHA256:
# DHE still gives of PFS but at increased computation cost
0%      DHE-RSA-AES256-GCM-SHA384:
0%      DHE-DSS-AES256-GCM-SHA384:
0%      DHE-RSA-AES128-GCM-SHA256:
0%      DHE-DSS-AES128-GCM-SHA256:
# CBC versions of the PFS ciphers
0%      ECDHE-ECDSA-AES256-SHA384:
30.6% ECDHE-RSA-AES256-SHA384:
0%      ECDHE-ECDSA-AES128-SHA256:
0%      ECDHE-RSA-AES128-SHA256:
0%      DHE-RSA-AES256-SHA256:
0%      DHE-DSS-AES256-SHA256:
0%      DHE-RSA-AES128-SHA256:
0%      DHE-DSS-AES128-SHA256:
0%      DHE-RSA-AES256-SHA:
0%      DHE-DSS-AES256-SHA:
0%      DHE-RSA-AES128-SHA:
0%      DHE-DSS-AES128-SHA:
0.3%   AES256-GCM-SHA384:
0%      AES128-GCM-SHA256:
0.2%   AES256-SHA256:
0%      AES128-SHA256:
4.8%   AES256-SHA:
0.5%   AES128-SHA:
# 3DES as a last resort
0.9%   DES-CBC3-SHA

View raw message