Very thanks for Julien Pierre answer.
I use webload 6.0 to do full SSL handshakes,SSL session can't reuse,and key
size=1024bit.
Could you offer  us  the name of SSL hardware accelerator cards which are
often used(test in nss)?


"Julien Pierre" <[EMAIL PROTECTED]> ????
news:[EMAIL PROTECTED]
> Liu,
>
> liupeng wrote:
> > Hi,
> >    Does someone can provide NSS performance report?
> >    I use nss-3.9 to build my ssl server,and use webload 6.0 test my
> > server,the machine information as following:
> > server machine:
> > Turbolinux 2.4.18 SMP
> > System configuration:
> > 2 Xeon 2.4G 512kb cache
> > 2G memory
> >
> > client machine:win2000*4
> >
> > The performance of my server seems not good,when loadsize=50,only about
160
> > transaction per second,and cpu idle=0,can someone provide nss
performance
> > report or explain this?
> >
> > Thanks
> >
> > LiuPeng
> >
> >
> >
>
> I have done a lot of performance work on NSS for AOL & Sun over the
> years, but none of the results are public.
>
> That said, there are some factors I can discuss with you that you
> haven't mentioned and are very important parameters.
>
> 1) Are you doing full SSL handshakes with RSA every time ? Or SSL
> session reuse ?
>
> 2) If full handshake, with what key size ? 1024 bits or 2048 bits ?
>
> RSA is still a very expensive computation, even for modern CPUs. It
> represents over 95% of the CPU cycles that NSS spends in SSL full
> handshakes. If you are doing full RSA, even at 1024 bits, then 160 ops/s
> seems a good number to get for the kind of server hardware you have.
>
> To get quite a bit more than that, you would have to buy something with
> more processors, with SSL hardware accelerator cards, or more likely
> with both.
>
> Note that NSS has been optimized for multithreading and multiple CPUs .
> Most of the server-side performance work on NSS has focused on Sun
> Ultrasparc CPU, not on the x86 architecture.
>
> Also, note that the Linux 2.4 and prior kernel is not very as good at
> multithreading and SMP as, say, Solaris is. I expect 2.6 to be better
> but I haven't had a chance to verify that.
>
> PS. I do work for Sun, so I have a visible bias, but I must tell you
> that I make those statements about x86 and Linux vs Sparc & Solaris
> because those are the facts I observed in my tests.


_______________________________________________
mozilla-crypto mailing list
[EMAIL PROTECTED]
http://mail.mozilla.org/listinfo/mozilla-crypto

Reply via email to