topic After I replaced malloc to in Intel® Integrated Performance Primitives

Is sgx_aes_ctr reentrant?

Joongun_P_ — Thu, 11 May 2017 10:53:39 GMT

In my enclave.cpp code, there is sgx_aes_ctr function calls.

When I executes my program, it works well but the execution time is not proportional to the 1 / (number of threads).

I tested with 7700k cpu and 1, 2, 4 threads.

Without sgx_aes_ctr calls, execution results are proportional to 4, 2, 1 times well.

I am using sgx sdk version 1.8 and I realize that the fuction is lapping of ippsAESEncryptCTR.

https://github.com/01org/linux-sgx/blob/1115c195cd60d5ab2b80c12d07e21663e5aa8030/sdk/tlibcrypto/sgx_aes_ctr.cpp

https://software.intel.com/sites/default/files/managed/f6/6f/ippcp_0.pdf

Is sgx_aes_ctr reentrant?

Why my programs execution time looks it doesn't?

It looks like you're not

Juan_d_Intel — Fri, 12 May 2017 13:02:39 GMT

It looks like you're not concerned about reentrancy but multithreading scalability, right?

Could you describe (or even better post) your code here showing how you measure sgx_aes_ctr and ippsAESEncryptCTR execution times?

If your program is calling sgx_aes_ctr multiple times, the lack of scalability might be caused by the malloc/free calls. Since you're using the 1.8 Linux SGX SDK, you could use the tcmalloc library, instead of the default memory allocator included in sgx_tstdc. If that's not the reason, I'd like to see the test program to better understand what the issue might be.

After I replaced malloc to

Joongun_P_ — Mon, 15 May 2017 03:20:50 GMT

After I replaced malloc to tcmalloc, execution times became scalable.

Thank you, Juan!