AES-SIV encryption/decryption has bad performance on small input and when run in many threads in parallel. #36
Assignees
Comments
copybara-service bot
pushed a commit
that referenced
this issue
Apr 29, 2024
We already do the same thing for other key type, such as AES-GCM. See also: #36, and #24. I run some benchmarks for this, where I encrypt some data using AES-SIV many times. - For 1kb data, it get 2.5x faster for a single thread, and 5.6x faster for 10 threads in parallel. - For 32 bytes data, it gets 15x faster for a single thread, and 41x faster for 10 threads in parallel. PiperOrigin-RevId: 629075782 Change-Id: I6735eafe4670213af15084343a7f269885d1101b
|
This is now fixed in the new release 1.14, https://github.com/tink-crypto/tink-java/releases/tag/v1.14.0. This release also contains other performance improvements for AES-SIV, and AES-CMAC, for example 862ecfc and ff90fb9. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This is the same issue as already discussed here for AES-GCM:
tink-crypto/tink#208
AES-SIV currently creates a new cipher object on each call to encryptDeterministically and decryptDeterministically. But creating these objects is expensive, and it requires locks, which make it even worse when done in multiple threads.
AES-SIV is often used to encrypt small values, for example EncryptedSharedPreferences uses it to encrypt keys in key-value pairs, which are typically small.
So we should add the same solution as proposed in tink-crypto/tink#208 and use ThreadLocal ciphers objects.
The text was updated successfully, but these errors were encountered: