eap Mailing List: Re: draft-urien-eap-smartcard-06.txt

[Mohamad Badra (badra]

Thomas Otto wrote:

> More precisely, the client needs almost 30 seconds to send
> ClientKeyExchange, CertificateVerify, and ChangeCipherSpec 
> (in frame 52), where I remember ClientKeyExchange requires 
> the 48 byte premaster secret. (which requires expensive 
> computations ..)

It is FULL TLS. That means you will not see all these messages in 
TLS-PSK :).
The high latency here is because of the PRF (see later)

> So, is there hope for a speedup or is EAP-TLS not suitable for the 
> EAP smartcard? How much Java influences this duration?
>
> In contrast to this, as answer to "Are smartcards performances 
> sufficient ?" we get the information, "Usually smart
> cards include crypto-processors that compute the RSA 2048 bits
> algorithm in less than 0,5s." 
>
> This is great, so I suppose the card above lacks of such a
> crypto-processor ? 

The card used in our test needs between 100 and 500 ms for RSA operations.
When you use a card handle SHA1 and MD5, the PRF will be negligeable. 
Take the example: RSA sign 342 bytes/s and verify 3287 [Rescorla] 
whereas SHA1 do 22838 Kbytes/s.

> May I compare this to the running time of my implementation of 
> EAP-PSK, which took for a full authentication only 0.75 seconds.
> There is a strange 0.25 second delay between Identity Request and 
> Response, so essentially we have 0,5 seconds for the protocol.
> By the way, the notebook used for this test was nothing special, just a
> 4year old 400 MHz machine ;-)

Please, when you compare 4year old 400 MHz machine, do it but not with a 
VERY weak support like the smart card that has in our test the following 
properties: CPU 8 bits, RAM 2304 bytes, ROM 96 Kbytes, E2PROM 34 Kbytes, 
Max. Clock 8 MHz, Max Data Rate 424 kbit/s.

Add to that:

1) The weak smart card does not have also SHA1 and MD5 co-processor and 
thus the PRF-TLS representes about 25% from the global performance. 
Another 15% for the digest and 12 to 28% for data transfert. More info 
are available at 
http://www.infres.enst.fr/~urien/security/aswn2004-urien.pdf

2) You compare the EAP-PSK to FULL EAP-TLS, but you forget to do it with 
EAP-TLS-PSK.
Rescorla used in SSL and TLS book a Pentium II 400 for benchmarkink TLS, 
[machine equivalentto your one :)]. The Resumed Handshake (The 
equivalent of TLS-PSK) takes between 0.0028 and 0.0038 seconds (see 
Rescorla, Page 213, Figure 6.26).

Badra