Benchmarks/PrivateNearestNeighborsSearchBenchmark/PrivateNearestNeighborsSearchBenchmark.swift

@@ -260,7 +260,7 @@ struct PnnsBenchmarkContext<Scheme: HeScheme> {
             .map { encryptionParams in try Context(encryptionParameters: encryptionParams) }
         self.processedDatabase = try database.process(config: serverConfig, contexts: contexts)
         self.client = try Client(config: clientConfig, contexts: contexts)
-        self.server = try Server(database: processedDatabase, config: serverConfig)
+        self.server = try Server(database: processedDatabase)
         self.secretKey = try client.generateSecretKey()
         self.evaluationKey = try client.generateEvaluationKey(using: secretKey)
 

Sources/PIRGenerateDatabase/GenerateDatabase.swift

@@ -73,10 +73,13 @@ struct GenerateDatabaseCommand: ParsableCommand {
 
     @Option var valueType: ValueTypeArguments
 
+    @Option(help: "The first keyword")
+    var firstKeyword: Int = 0
+
     mutating func run() throws {
         let databaseRows = (0..<rowCount)
             .map { rowIndex in
-                let keyword = [UInt8](String(rowIndex).utf8)
+                let keyword = [UInt8](String(firstKeyword + rowIndex).utf8)
                 guard let valueSize = valueSize.range.randomElement() else {
                     preconditionFailure("Could not sample valueSize from range \(valueSize.range)")
                 }

Sources/PIRGenerateDatabase/PIRGenerateDatabase.docc/PIRGenerateDatabase.md

@@ -23,11 +23,11 @@ The binary will be generated in `.build/release/PIRGenerateDatabase`.
 PIRGenerateDatabase \
     --output-database database.txtpb \
     --row-count 100 \
-    --value-size '10..<20' \
+    --value-size '10...20' \
     --value-type repeated
 ```
 
-This will generate a database of 100 rows, with keywords 0 to 100, and each value repeating the keyword for 10 to 20 bytes.
+This will generate a database of 100 rows, with keywords 0 to 99, and each value repeating the keyword for 10 to 20 bytes.
 
 The database is a serialized [Apple_SwiftHomomorphicEncryption_Pir_V1_KeywordDatabase](https://swiftpackageindex.com/apple/swift-homomorphic-encryption/main/documentation/privateinformationretrievalprotobuf/apple_swifthomomorphicencryption_pir_v1_keyworddatabase).
 For readability, the `.txtpb` extension ensures the output database will be saved in protocol buffer text format.

Sources/PIRProcessDatabase/PIRProcessDatabase.docc/ReusingPirParameters.md

@@ -0,0 +1,126 @@
+# Reusing PIR Parameters
+
+Learn how to reuse PIR parameters across database updates.
+
+## Overview
+Database updates may lead to different PIR configurations, which need to be synced to the client.
+We can configure the database processing to yield the same PIR configuration across database updates.
+
+### Requirements
+This example assumes that you have the following binaries available on your `$PATH`.
+The binaries are:
+ - `PIRGenerateDatabase`
+ - `PIRProcessDatabase`
+
+The way to add these to your path is by first making sure that the `~/.swiftpm/bin` directory is on your `$PATH`. To do
+so, add the following line to your `~/.zshrc` or appropriate shell configuration file.
+```sh
+export PATH="$HOME/.swiftpm/bin:$PATH"
+```
+Make sure to reload it (`source ~/.zshrc`) or by restarting your terminal emulator. Then we are going to use the
+`experimental-install` feature of Swift Package Manager.
+
+Change directory to a checkout of this repository and run the following command.
+```sh
+swift package experimental-install -c release --product PIRGenerateDatabase
+swift package experimental-install -c release --product PIRProcessDatabase
+```
+
+### Example
+#### Database Creation
+
+1. We start by generating a sample database.
+
+```sh
+PIRGenerateDatabase \
+    --output-database /tmp/database-v1.txtpb \
+    --row-count 10000 \
+    --value-size '10...20' \
+    --value-type repeated
+```
+
+This will generate a database of 10000 rows, with keywords 0 to 9999, and each value repeating the keyword for 10 to 20 bytes.
+
+We view a few rows from the database with
+```sh
+head /tmp/database-v1.txtpb
+```
+which shows
+```json
+rows {
+  keyword: "0"
+  value: "000000000000000"
+}
+rows {
+  keyword: "1"
+  value: "111111111111111111"
+}
+rows {
+  keyword: "2"
+```
+
+2. To simulate an updated database, we generate a database update with 1000 more rows and different value sizes.
+```
+PIRGenerateDatabase \
+    --output-database /tmp/database-update.txtpb \
+    --row-count 1000 \
+    --first-keyword 10000 \
+    --value-size '10..<25' \
+    --value-type repeated
+```
+
+We combine the initial database with the update to create `/tmp/database-v2.txtpb`.
+```sh
+cat /tmp/database-v1.txtpb /tmp/database-update.txtpb > /tmp/database-v2.txtpb
+```
+
+#### Processing
+To ensure processing the update database yields the same configuration, we use the `.fixedSize` cuckoo table argument, specifying a bucket count.
+A larger bucket count will leave more room for new entries, without changing the configuration.
+However, a larger bucket count will also increase server runtime.
+One way to choose the `bucketCount` is to start with `bucketCount : 1` and try larger `bucketCounts` until the processing works.
+If the processing throws a `PirError.failedToConstructCuckooTable` or logs `Failed to construct Cuckoo table`, this is an indication the chosen bucket count was too small.
+
+We create `/tmp/config-v1-fixed-size.json` with the following contents
+```
+{
+  "algorithm" : "mulPir",
+  "cuckooTableArguments" : {
+    "bucketCount" : {
+      "fixedSize" : {
+        "bucketCount" : 256,
+      }
+    },
+    "hashFunctionCount" : 2,
+    "maxEvictionCount" : 100,
+    "maxSerializedBucketSize" : 1024
+  },
+  "inputDatabase" : "/tmp/database-v1.txtpb",
+  "keyCompression" : "noCompression",
+  "outputDatabase" : "/tmp/database-v1-SHARD_ID.bin",
+  "outputEvaluationKeyConfig" : "/tmp/database-v1-evaluation-key-config.txtpb",
+  "outputPirParameters" : "/tmp/database-v1-pir-parameters-SHARD_ID.txtpb",
+  "rlweParameters" : "n_4096_logq_27_28_28_logt_5",
+  "sharding" : {
+    "shardCount" : 2
+  },
+  "trialsPerShard" : 1
+}
+```
+and process the original database with
+```sh
+PIRProcessDatabase /tmp/config-v1-fixed-size.json
+```
+
+We can get a hash of the output parameters with `cat /tmp/database-v1-pir-parameters-*.txtpb | shasum`
+
+We create `/tmp/config-v2-fixed-size.json` by replacing `v1` with `v2`:
+```sh
+sed 's/v1/v2/g' /tmp/config-v1-fixed-size.json > /tmp/config-v2-fixed-size.json
+```
+and process the updated database with
+```sh
+PIRProcessDatabase /tmp/config-v2-fixed-size.json
+```
+We can get a hash of the output parameters with `cat /tmp/database-v2-pir-parameters-*.txtpb | shasum`.
+This should match the shasum from `cat /tmp/database-v1-pir-parameters-*.txtpb | shasum`.

Sources/PrivateInformationRetrieval/CuckooTable.swift

@@ -347,7 +347,7 @@ struct CuckooTable {
                         """
                         Unable to insert value with keyword \(keywordValuePair.keyword) \
                         into table with \(entryCount) entries. \
-                        Consider enabling table expansion in config.
+                        Consider setting `allowExpansion` or increasing `bucketCount`.
                         """)
             }
         }

Sources/PrivateInformationRetrieval/PrivateInformationRetrieval.docc/ParameterTuning.md

@@ -1,6 +1,6 @@
 # Parameter Tuning
 
- Keyword PIR Parameter Tuning
+Learn how to tune parameters for Keyword PIR.
 
 ## Overview
 
@@ -68,7 +68,7 @@ executables.
 PIRGenerateDatabase \
     --output-database thinDatabase.txtpb \
     --row-count 100000 \
-    --value-size '2' \
+    --value-size 2 \
     --value-type random
 ```
 

Sources/PrivateInformationRetrieval/PrivateInformationRetrieval.docc/PrivateInformationRetrieval.md

@@ -22,3 +22,4 @@ The PIR implementation in Swift Homomorphic Encryption uses homomorphic encrypti
 ### Articles
 - <doc:EncodingPipeline>
 - <doc:ParameterTuning>
+- <doc:ReusingPirParameters>

Sources/PrivateInformationRetrieval/PrivateInformationRetrieval.docc/ReusingPirParameters.md

@@ -0,0 +1,134 @@
+# Reusing PIR Parameters
+
+Reusing PIR parameters across database updates.
+
+## Overview
+Database updates may lead to different PIR configurations, which need to be synced to the client.
+We can configure the database processing to yield the same PIR configuration on small database updates.
+
+### Requirements
+This example assumes that you have the following binaries available on your `$PATH`.
+The binaries are:
+ - `PIRGenerateDatabase`
+ - `PIRProcessDatabase`
+
+The way to add these to your path is by first making sure that the `~/.swiftpm/bin` directory is on your `$PATH`. To do
+so, add the following line to your `~/.zshrc` or appropriate shell configuration file.
+```sh
+export PATH="$HOME/.swiftpm/bin:$PATH"
+```
+Make sure to reload it (`source ~/.zshrc`) or by restarting your terminal emulator. Then we are going to use the
+`experimental-install` feature of Swift Package Manager.
+
+Change directory to a checkout of this repository and run the following command.
+```sh
+swift package experimental-install -c release --product PIRGenerateDatabase
+swift package experimental-install -c release --product PIRProcessDatabase
+```
+
+### Example
+#### Database Creation
+
+1. We start by generating a sample database.
+
+```sh
+PIRGenerateDatabase \
+    --output-database /tmp/database-v1.txtpb \
+    --row-count 10000 \
+    --value-size '10...20' \
+    --value-type repeated
+```
+
+This will generate a database of 100 rows, with keywords 0 to 99, and each value repeating the keyword for 10 to 20 bytes.
+
+We view a few rows from the database with
+```sh
+head /tmp/database-v1.txtpb
+```
+which shows
+```json
+rows {
+  keyword: "0"
+  value: "000000000000000"
+}
+rows {
+  keyword: "1"
+  value: "111111111111111111"
+}
+rows {
+  keyword: "2"
+```
+
+2. To simulate an updated database, we generate a database update with a few more rows and different value sizes
+```
+PIRGenerateDatabase \
+    --output-database /tmp/database-update.txtpb \
+    --row-count 1000 \
+    --first-keyword 10000 \
+    --value-size '10..<25' \
+    --value-type repeated
+```
+
+We combine the initial database with the update to create `/tmp/database-v2.txtpb`.
+```sh
+cat /tmp/database-v1.txtpb /tmp/database-update.txtpb > /tmp/database-v2.txtpb
+```
+
+#### Processing
+To ensure processing the update database yields the same configuration, we use the `.fixedSize` cuckoo table argument, specifying a bucket count.
+A larger bucket count will leave more room for new entries, without changing the configuration.
+However, a larger bucket count will also increase server runtime.
+One way to choose the `bucketCount` is to start with `bucketCount : 1` and try larger `bucketCounts` until the processing succeeds.
+If the processing throws a `PirError.failedToConstructCuckooTable` or logs `Failed to construct Cuckoo table`, this is an indication the chosen bucket count was too small.
+
+> Important: We also specify a fixed shard count, rather than using `entryCountPerShard`.
+
+We create `/tmp/config-v1-fixed-size.json` with the following contents
+```json
+{
+  "algorithm" : "mulPir",
+  "cuckooTableArguments" : {
+    "bucketCount" : {
+      "fixedSize" : {
+        "bucketCount" : 256,
+      }
+    },
+    "hashFunctionCount" : 2,
+    "maxEvictionCount" : 100,
+    "maxSerializedBucketSize" : 1024
+  },
+  "inputDatabase" : "/tmp/database-v1.txtpb",
+  "keyCompression" : "noCompression",
+  "outputDatabase" : "/tmp/database-v1-SHARD_ID.bin",
+  "outputEvaluationKeyConfig" : "/tmp/database-v1-evaluation-key-config.txtpb",
+  "outputPirParameters" : "/tmp/database-v1-pir-parameters-SHARD_ID.txtpb",
+  "rlweParameters" : "n_4096_logq_27_28_28_logt_5",
+  "sharding" : {
+    "shardCount" : 2
+  },
+  "trialsPerShard" : 1
+}
+```
+and process the original database with
+```sh
+PIRProcessDatabase /tmp/config-v1-fixed-size.json
+```
+
+We hash the output PIR parameters with
+```sh
+cat /tmp/database-v1-pir-parameters-*.txtpb | shasum
+```
+
+We create `/tmp/config-v2-fixed-size.json` by replacing `v1` with `v2`:
+```sh
+sed 's/v1/v2/g' /tmp/config-v1-fixed-size.json > /tmp/config-v2-fixed-size.json
+```
+and process the updated database with
+```sh
+PIRProcessDatabase /tmp/config-v2-fixed-size.json
+```
+We hash the output PIR parameters of the updated database with
+```sh
+cat /tmp/database-v2-pir-parameters-*.txtpb | shasum
+```
+This should match the hash of the original database's PIR parameters.

Sources/PrivateNearestNeighborsSearch/Server.swift

@@ -16,12 +16,14 @@ import HomomorphicEncryption
 
 /// Private nearest neighbors server.
 public struct Server<Scheme: HeScheme> {
-    /// Configuration.
-    public let config: ServerConfig<Scheme>
-
     /// The database.
     public let database: ProcessedDatabase<Scheme>
 
+    /// Configuration.
+    public var config: ServerConfig<Scheme> {
+        database.serverConfig
+    }
+
     /// Configuration needed for private nearest neighbors search..
     public var evaluationKeyConfiguration: EvaluationKeyConfiguration {
         config.clientConfig.evaluationKeyConfig
@@ -33,16 +35,13 @@ public struct Server<Scheme: HeScheme> {
     }
 
     /// Creates a new ``Server``.
-    /// - Parameters:
-    ///   - database: Processed database.
-    ///   - config: Server configuration.
+    /// - Parameter database: Processed database.
     /// - Throws: Error upon failure to create the server.
     @inlinable
-    public init(database: ProcessedDatabase<Scheme>, config: ServerConfig<Scheme>) throws {
-        guard config.distanceMetric == .cosineSimilarity else {
-            throw PnnsError.wrongDistanceMetric(got: config.distanceMetric, expected: .cosineSimilarity)
+    public init(database: ProcessedDatabase<Scheme>) throws {
+        guard database.serverConfig.distanceMetric == .cosineSimilarity else {
+            throw PnnsError.wrongDistanceMetric(got: database.serverConfig.distanceMetric, expected: .cosineSimilarity)
         }
-        self.config = config
         self.database = database
     }
 

Tests/PrivateNearestNeighborsSearchTests/ClientTests.swift

@@ -201,7 +201,7 @@ final class ClientTests: XCTestCase {
             let processed = try database.process(config: serverConfig)
 
             let client = try Client(config: clientConfig, contexts: processed.contexts)
-            let server = try Server(database: processed, config: serverConfig)
+            let server = try Server(database: processed)
 
             // We query exact matches from rows in the database
             let queryVectors = Array2d(data: database.rows.prefix(queryCount).map { row in row.vector })