I've been wondering how and when OpenSSL decides whether it can use the new aes instructions? Does it decide at build time or at run time? If I build on a CPU that supports aes instructions but run on a cpu that does not, will bad things happen? Or is OpenSSL smart enough to call functions implemented without aes instructions in that case? Norm Green On 8/10/16 06:28, Jan Just Keijser wrote: > Hi, > > On 10/08/16 14:25, Nagesh shamnur wrote: >> >> Hi Group, >> >> I am running an application which transfers huge chunks of data every >> second (850Mbps) and the same is secured using openssl. However the >> CPU usage on windows is very high ( ~ 100%). So as a part of the >> analysis, I stumbled upon the information that, when using AES >> encryption, if the underlying hardware is Intel CPU, it can support >> AES-NI instruction set and hence make the crypto processing faster. >> So, I wanted to confirm if the same is enabled in my hardware. >> >> So, I wanted to know how to verify if the run is able to use the >> AES-NI instruction set available in the hardware. >> >> I have built openssl and have ensured enabling the asm in both linux >> and windows build. >> >> For windows, to confirm if AES-NI is enabled, support of tools >> available like truecrypt, CPU-Z and blackbox were used if the same >> was enabled in OS usage. And I found that the same is disabled. Also >> I found in some blogs that the same needs to be enabled in BIOS. When >> checked the BIOS settings, the option was not be found and a BIOS >> update is required to enable the same. >> >> However in linux I was unable to conclude if AES-NI is disabled since >> I didn?t had access to any such tools on linux. I checked "#cpuinfo | >> grep aes" and i was unable to find any line regarding AES-NI. However >> when i run the ./openssl speed -evp aes-128-gcm and >> OPENSSL_ia32cap="~0x200000200000000" ./openssl speed -elapsed -evp >> aes-128-gcm i am able to find the difference in speed. So i wanted to >> check how to confirm if my linux build has AES-NI enabled or not? >> >> Environment Information: >> >> CPU: E5-2620 0 @2.0GHz >> >> OS: Windows Server 2008 >> >> Linux: Ubuntu 3.11.0-15-generic >> >> Openssl versoin: 1.0.2h >> >> > I've got a server with that exact same CPU over here; with openssl > 1.0.2d I see the following results: > > $ ./openssl speed -evp aes-128-gcm > [...] > type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 > bytes > aes-128-gcm 184391.41k 465791.06k 689190.61k .65k 781295.62k > > $ OPENSSL_ia32cap=0 ./openssl speed -evp aes-128-gcm > [...] > type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 > bytes > aes-128-gcm 43906.03k 49490.24k 51037.70k 51554.65k > 51699.71k > > i.e. with AES-NI disabled performance is about ~15 times less. On this > CPU turboboost is not working so your numbers maybe slightly different. > Another good way to test whether AES-NI is working is by comparing > BF-CBC to AES-256-CBC: without AES-NI, BF will be faster. with AES-NI, > AES will be faster. > > HTH, > > JJK > > > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mta.openssl.org/pipermail/openssl-users/attachments/20160810/b33f0b70/attachment-0001.html>