Hello,
I am just doing some IPsec performance tests on shiny new DL 380 G8 (CPU is
Intel(R) Xeon(R) CPU E5-2643 @ 3.30GHz).
Here is the setup :
Two Optiplex <-> HP DL380 G8 <-> HP DL 380 G8 <-> Two Optiplex
Intel Gb NIC in every computer
All running 5.2-beta amd64 compiled yesterday
2x1gb trunk between the DL 380 g8
Max throughput is measured with tcpbench -n 10, max PPS is with
tcpbench -u
-B 64
I got those results :
Unencrypted :
Max thoughput : 1800 Mbps
Max PPS : 250 kpps
AES 256 :
Max thoughput : 410 Mbps
Max PPS : 70 kpps
AES 256 GCM :
Max thoughput : 570 Mbps
Max PPS : 95 kpps
AES 128 :
Max thoughput : 430 Mbps
Max PPS : 75 kpps
AES 128 GCM :
Max thoughput : 575 Mbps
Max PPS : 85 kpps
--
Cordialement,
Pierre BARDOU
-----Message d'origine-----
De : Mike Belopuhov [mailto:[email protected]]
Envoyé : mardi 26 juin 2012 14:39
À : Mark Romer
Cc : Ted Unangst; [email protected]; Ryan McBride
Objet : Re: ipsec tunnel speeds
On Mon, Jun 25, 2012 at 2:53 PM, Mark Romer <[email protected]> wrote:
> Great question Ted
> Does anyone know the answer?
sure.
> Thanks Mark
> On Jun 22, 2012 12:58 PM, "Ted Unangst" <[email protected]> wrote:
>
>> On Fri, Jun 22, 2012 at 12:52, Ryan McBride wrote:
>>
>> > 550Mb/s with aes-128-gcm (requires AES-NI and amd64) on
>> > hw.model=Intel(R) Xeon(R) CPU E5649 @ 2.53GHz hw.vendor=HP
>> > hw.product=ProLiant DL360 G7
>>
>> what's the reason aes-128-gcm requires amd64?
because the assembly is written for amd64.
>> we can't add that code to i386?
that specific one? of course not. but the aes-ni and clmul instructions are
part of sse and can be executed by both 32-bit and 64-bit programs.
apart from that, it might be possible that binutils have to be adjusted (i
don't remember if they share the same code) and i386 has to grow
fpu_kernel_{enter,exit}.
