Re: [Ntp] Follow-up to yesterday's mic comment about PTP security Wed, 24 July 2019 07:00 UTC

Return-Path: <>
Received: from localhost (localhost []) by (Postfix) with ESMTP id A1C191202DF for <>; Wed, 24 Jul 2019 00:00:12 -0700 (PDT)
X-Virus-Scanned: amavisd-new at
X-Spam-Flag: NO
X-Spam-Score: -1.899
X-Spam-Status: No, score=-1.899 tagged_above=-999 required=5 tests=[BAYES_00=-1.9, HTML_MESSAGE=0.001, RCVD_IN_DNSWL_NONE=-0.0001, SPF_HELO_NONE=0.001, SPF_PASS=-0.001] autolearn=ham autolearn_force=no
Received: from ([]) by localhost ( []) (amavisd-new, port 10024) with ESMTP id 5x1-E-ucsvMd for <>; Wed, 24 Jul 2019 00:00:10 -0700 (PDT)
Received: from ( []) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by (Postfix) with ESMTPS id C0F31120332 for <>; Wed, 24 Jul 2019 00:00:09 -0700 (PDT)
Received: from ( []) by with ESMTP id x6O707PN008752-x6O707PP008752 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=FAIL); Wed, 24 Jul 2019 09:00:07 +0200
Received: from ( []) by (Postfix) with ESMTPS id 19DD58144E0; Wed, 24 Jul 2019 09:00:07 +0200 (CEST)
In-Reply-To: <>
References: <>
To: "Daniel Franke" <>
Cc: "NTP WG" <>
MIME-Version: 1.0
Message-ID: <>
Date: Wed, 24 Jul 2019 09:00:54 +0200
Content-Type: multipart/alternative; boundary="=_alternative 002675FCC1258441_="
Archived-At: <>
Subject: Re: [Ntp] Follow-up to yesterday's mic comment about PTP security
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: <>
List-Unsubscribe: <>, <>
List-Archive: <>
List-Post: <>
List-Help: <>
List-Subscribe: <>, <>
X-List-Received-Date: Wed, 24 Jul 2019 07:00:13 -0000

Hey all,

first of all, I'm really glad if this whole thing (security of one-way 
mechanisms and mechanism selection) is a discussion that we're going to 
have in the WG.

To comment on your assertions, Daniel:

It is established in general (and I have a proof lying around for a model 
of NTS in particular) that a client performing a request-response exchange 
with NTS and using all relevant checks gets a strong guarantee that the 
error in its measured offset is no larger than half the added flight times 
of the packets (plus some negligibly small delta accounting for frequency 
instability of the clocks used on client and server side). 
For anyone wondering why we bothered to prove this again: this guarantee 
is 100%, and the new part is "no matter what a Man-in-the-Middle attacker 
did in the process".
So I would be careful about naming a specific amount, because flight times 
do depend on the specific client's connection - but 50ms seems like a good 
rule of thumb, and I overall agree with your assertion.

If we're operating und the assumptions that 
a) you can only use one time sync mechanism at a time and keep track of 
one clock disciplined via data from that mechansim, and
b) end users always have exactly one requirement level for each of 
security and precision/accuracy and need to use the least-effort path to 
achieve them
... then I agree with your assertions whole-heartedly.

But I really think both assumptions deserve their own hard looks and 
For example, it might be reasonable for someone, specifically a financial 
institute, to run NTP with NTS in their local network to obtain a 100% 
security guarantee for a 100us level (demanded by MiFID II for example) 
and also still use PTP / White Rabbit (unsecured for the time being) to 
have the precision/accuracy levels they actually want - with no strong 
guarantee, but still valid in the (most likely) case that their 
infrastructures are not currently under attack.

Again, I agree with the assertion for the most part and in the given 
status quo.
But the underlying assumption that every relevant adversarial 2-way 
network also suffers from long, unpredictable and asymmetric travel times 
is mostly valid because the only candidate for such a network is the 
If someone built, say, a GNSS network where two-way communication (with 
satellites or between two ground stations) was readily available to 
everyone, the whole situation would be different:
That would still potentially qualify as an adversarial network, but with 
the proper crypto, your 100% security guarantees could be extended to much 
better precision/accuracy levels.
The same thing could be true for long-distance tree-topology fibre-based 
networks exclusively for time synchronization - which are kind of in the 
process of being built all over Europe.

Lengthy comments and caveats notwithstanding, I agree with and would 
endorse your 1.-4. decision making sheet as an excellent starting point.

Best regards,

"ntp" <>; schrieb am 23.07.2019 18:19:33:

> Von: "Daniel Franke" <>;
> An: "NTP WG" <>;
> Datum: 23.07.2019 18:20
> Betreff: [Ntp] Follow-up to yesterday's mic comment about PTP security
> Gesendet von: "ntp" <>;
> My comments yesterday about PTP security shifted context a few times
> so it may have been hard to follow what I was claiming. My assertions
> were:
> 1. If you need 50ms precision, pick some good public NTP servers and use 
> 2. If you need 100µs precision, colocate a time source in the same
> datacenter as the client systems. Use NTP and NTS; you don't need PTP
> for this.
> 3. If you need 1µs precision, use PTP and physically secure the link
> between the time source and the clients so that cryptographic
> authentication is unnecessary.
> 4. If you need 1µs precision over an adversarial network, good luck!
> This is simply not achievable and no amount of cryptographic pixie
> dust is ever going to save you.
> _______________________________________________
> ntp mailing list