## Question 4
There are two test classes `CartTest` and `SupermarketTest` to test for `Cart` and `Supermarket` both for local test and RPC test. The `localTest` indicates which kind of test is performed. There are two kind of tests. One kind is for testing the basic correctness of implementation in single thread. It tests both the normal cases and cases where exceptions are expected to be thrown.
Another is to test before-or-after atomicity. In the test `testForConcurrency1` in `SupermarketTest` , `thread 1` repeatedly adds items to one cart and check out which will update the corresponding items’ stock in the supermarket. `thread 2` repeatedly calls the `updateStocks` to replenish the corresponding items. The amount of stocks `thread 1 ` consumes eqaul to the amount `thread 2` increases. The test method runs the two threads at the same time, when they finish, assert that the amount of stocks don’t change. In the test `testForConcurrency2` in `SupermarketTest` , `thread 1` repeatedly call `updateStocks` that increase the stocks and then `updateStocks` that decreases the stocks , `thread 2` repeatedly call `getItems`, assert that the `getItems` returns the amount of stocks either after decrease or after increase. In the test `testForConcurrency1` in `CartTest` , `thread 1` repeatedly adds items to one cart . `thread 2` repeatedly remove the same items from the same cart. The amount of items `thread 1 `adds eqaul to the amount `thread 2` removes. The test method runs the two threads at the same time, when they finish, assert that the quantity of the item doesn’t change. In the test `testForConcurrency2` in `CartTest` , `thread 1` repeatedly call `add` and then `remove` , `thread 2` repeatedly call `getCartItems`, assert that the `getCartItems` returns the quantity of cart item either after `add` or after `remove`.
## Question 5
### Workload Design
There are two kinds of access to `Supermarket` service: one is by `Cart` and another is by `Admin Client`. For `Cart` , the `getItems` call is invoked when `Stock Client` add a `Cart Item` to the `Cart` and `updateStocks` is invoked when `Mobile Client` checkout. Assuming on average, custom add 10 different items before checkout. Thus in `runCartInteraction`, it calls `updateStocks` with probability of 0.1 and calls `getItems` with probability 0.9. For `Admin Client` , it mainly uses `updateStocks` to replenish items. It calls `updateStocks` with probability 0.8 and calls `getItems` with probability 0.2. Since `Cart` interaction is much more frequent than `Admin Client` interaction. `runAdminClientInteraction` is called with probability 0.01 and `runCartInteraction` is called with probability 0.99.
### Experimental Setup
**Hardware**: MacBook Pro (15-inch, 2016) with processor 2.7 GHz Intel Core i7 and memory 16 GB 2133 MHz LPDDR3. With more powerful hardware, the throughput is expected to increase.
**Data**: 100000 items are generated in the supermarket initially.
**Configuration**: Use binary serialization in RPC. Worker with 5000 warm up runs and 20000 actual runs. Max thread pool set to 100 and min thread pool set to 10 in `SupermarketHTTPServer`.
Each worker thread records the successful interactions and the time taken. Total successful interactions divided by total time taken to get the throughput.
### Discussion of Results
Above graph shows the throughput changes with number of client threads. We can see that the throughput tends to decrease when we increase the number of clients. This maches my expectations. More clients will compete for the computation resources, and each client will take more time, thus the throughput decreases.