porting to SEAL 3.7.2

README.md

@@ -2,52 +2,82 @@
 
 SealPIR is a research library and should not be used in production systems. 
 SealPIR allows a client to download an element from a database stored by a server without
-revealing which element was downloaded. SealPIR was introduced at 
-the Symposium on Security and Privacy (Oakland) in 2018. You can find
+revealing to the server which element was downloaded. SealPIR was introduced at 
+the IEEE Symposium on Security and Privacy (Oakland) in 2018. You can find
 a copy of the paper [here](https://eprint.iacr.org/2017/1142.pdf).
 
-This is a newer version of SealPIR that uses the latest version of SEAL
-and provides better serialization/deserialization of queries and responses,
-and a more streamlined code base. A drawback of this version is that ciphertexts 
-are slightly larger (due to specifics with SEAL).
-
-If you wish to use the **original** version of SealPIR which uses an older version 
-of SEAL and **has smaller ciphertexts**, check out the 
-[original](https://github.com/microsoft/SealPIR/tree/ccf86c50fd3291) branch in this 
-repository.
-
 # Compiling SEAL
 
-SealPIR depends on [Microsoft SEAL version 3.6.6](https://github.com/microsoft/SEAL/tree/3.6.6).
-Install SEAL before compiling SealPIR.
+SealPIR depends on [Microsoft SEAL version 3.7.2](https://github.com/microsoft/SEAL/tree/3.7.2).
+
+Download and install SEAL (follow the instructions in the above link) before before compiling SealPIR.
 
 # Compiling SealPIR
 
-Once Microsoft SEAL 3.6.6 is installed, to build SealPIR simply run:
+Once Microsoft SEAL 3.7.2 is installed, to build SealPIR simply run:
 
 ```
 cmake .
 make
 ```
-	
+
 This should produce a binary file ``bin/main``.
 
 # Testing SealPIR
 
-Once you have compiled SealPIR, you can test that everything is working
-correctly by running:
+Once you have compiled SealPIR, you can run our battery of unit tests with:
 
 ```
-ctest
+ctest .
 ```
 
 # Using SealPIR
 
 Take a look at the example in `src/main.cpp` for how to use SealPIR. 
 You can also look at the tests in the `test` folder.
-Note: the parameter "d" stands for recursion levels, and for the current 
-configuration, the server-to-client reply has size (pow(10, d-1) * 32) KB. 
-Therefore we recommend using d <= 3.  
+
+
+## Default parameters
+
+$$N$$ indicates the degree of the BFV polynomials.  Default is 4096.
+
+$$t$$ indicates the plaintext modulus, but we specify $$logt$$ instead. Default is 20.
+
+Each BFV ciphertext can encrypt $$\log{t}\cdot N \approx 10 KB$$ bits of information.
+
+This means that if your database has, say, 1 KB elements, then you can pack 10 
+such elements into a single BFV plaintext. 
+On the other hand, if your database has, say, 20 KB elements, then you will 
+need two BFV plaintexts to represent each of your elements.
+
+$$d$$ represents the recursion level.  When the number of BFV plaintexts needed
+to represent your database (see above for how to map the number of database
+elements of a given size to the number of BFV plaintexts) is smaller than N,
+then setting $$d = 1$$ minimizes communication costs. However, you can also set
+$$d = 2$$ which doubles the size of the query and increases the size of the
+response by roughly a factor of 4, but in some cases might reduce computational
+costs a little bit (because the oblivious expansion procedure is cheaper). 
+
+When the number of BFV plaintexts is much greater than N, then $$d = 2$$
+minimizes communication costs. You can read the paper to understand how $$d$$
+affects communication costs. In general, the query consists of $$d$$ BFV
+ciphertexts and can index a database with $$N^d$$ BFV plaintexts;  the response
+consists of $$F^{d-1}$$ ciphertexts, where $$F$$ is the ciphertext
+expansion factor. In the current implementation which uses recursive
+modulo swithcing, $$F$$ is around 4. We have not identified any setting where
+$$d > 2$$ is beneficial.
+
+
+# Changelog
+
+This implementation of SealPIR uses the latest version of SEAL, fixes several bugs,
+and provides better serialization/deserialization of queries and responses,
+and a more streamlined code base.
+
+If you wish to use the **original** version of SealPIR which corresponds to the
+numbers reported in the paper and which uses an older version  of SEAL, check
+out [this](https://github.com/microsoft/SealPIR/tree/ccf86c50fd3291) branch in
+the git repository.
 
 # Contributing
 

src/CMakeLists.txt

@@ -1,4 +1,4 @@
-find_package(SEAL 3.6 REQUIRED)
+find_package(SEAL 3.7.2 REQUIRED)
 
 add_library(sealpir pir.hpp pir.cpp pir_client.hpp pir_client.cpp pir_server.hpp
   pir_server.cpp)

src/main.cpp

@@ -14,7 +14,7 @@ using namespace seal;
 
 int main(int argc, char *argv[]) {
 
-    uint64_t number_of_items = 1 << 12;
+    uint64_t number_of_items = 1 << 16;
     uint64_t size_per_item = 256; // in bytes
     uint32_t N = 4096;