Common use of SCoin Clause in Contracts

SCoin. We now present results for the SCoin application defined in Section V-A. In these experiments, we benchmark how well the protocol can perform with a single application with varying number of transactions that require cross-shard communication (i.e., tokens transferred between different partitions). We try to measure latency and throughput tradeoffs with a varying number of cross-shard transactions. Each client in the experiment tries to execute transfer transactions in a closed-loop. If the transaction is cross-shard, i.e., if a client is trying to transfer its token to an account that resides in a different shard, the client first move its account to the corresponding shard and then executes the transfer transaction afterward in the destination shard. Similarly to the previous experiment, we tune the number of clients in the system to avoid a significant degradation of the average latency for each client, thus caping each shard to 250 clients. We experiment with different cross-shard transaction rates for different shard numbers. Figure 6 shows the aggregate throughput of the system for a varying number of shards and different percentages of cross-shard transactions. The one shard experiment is shown in every cross-shard rate experiment as a reference. We can see how performance degrades when increasing the number of cross-shard transaction rates, but the throughput grows linearly at different rates of cross-shard. For comparison, the previ- ous experiment had on average 5.86%, 7.93%, 7.85% cross- blockchain transaction rate for 2, 4 and 8 shards, respectively. Any system that reserves part of its allocated resources to process cross-shard transactions will have similar behavior, due to cross-shard transactions occupying the resources that otherwise would be used by single-shard transactions. The average latency for clients does not change significantly when increasing the number of shards, remaining at around 7 seconds for single-shard transactions and 34 seconds for cross-shard transactions. Cross-shard transactions demand two transactions for each move operation, plus waiting for two single-shard cross-shard aggregated single-shard cross-shard aggregated 1.0 1.0 0.8 0.8 % % 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 blocks to prove the contract’s state and one final transaction to complete the operation, confirming the expected latency of waiting for five blocks per cross-shard transaction. In Figure 7 (right) we can see the cumulative distribution function (CDF) for clients’ observed latencies in a scenario with four shards and 10% cross-shard transactions rate. The aggregated latency shows both single and cross-shard transactions and we can see that, as expected, around 10% of the transactions takes more than 30 seconds to complete. Differently from other sharded systems (e.g., [28]) the protocol does not suffer from a convoy effect [29], that is, cross-shard transactions do not delay single-shard transactions.

SCoin. We now present results for the SCoin application defined in Section V-A. In these experiments, we benchmark how well the protocol can perform with a single application with varying number of transactions that require cross-shard communication (i.e., tokens transferred between different partitions). We try to measure latency and throughput tradeoffs with a varying number of cross-shard transactions. Each client in the experiment tries to execute transfer transactions in a closed-loop. If the transaction is cross-shard, i.e., if a client is trying to transfer its token to an account that resides in a different shard, the client first move its account to the corresponding shard and then executes the transfer transaction afterward in the destination shard. Similarly to the previous experiment, we tune the number of clients in the system to avoid a significant degradation of the average latency for each client, thus caping each shard to 250 clients. We experiment with different cross-shard transaction rates for different shard numbers. Figure 6 shows the aggregate throughput of the system for a varying number of shards and different percentages of cross-shard transactions. The one shard experiment is shown in every cross-shard rate experiment as a reference. We can see how performance degrades when increasing the number of cross-shard transaction rates, but the throughput grows linearly at different rates of cross-shard. For comparison, the previ- ous experiment had on average 5.86%, 7.93%, 7.85% cross- blockchain transaction rate for 2, 4 and 8 shards, respectively. Any system that reserves part of its allocated resources to process cross-shard transactions will have similar behavior, due to cross-shard transactions occupying the resources that otherwise would be used by single-shard transactions. The average latency for clients does not change significantly when increasing the number of shards, remaining at around 7 seconds for single-shard transactions and 34 seconds for cross-shard transactions. Cross-shard transactions demand two transactions for each move operation, plus waiting for two single-shard cross-shard aggregated single-shard cross-shard aggregated 1.0 1.0 0.8 0.8 % % 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 blocks to prove the contract’s state and one final transaction to complete the operation, confirming the expected latency of waiting for five blocks per cross-shard transaction. In Figure 7 (right) we can see the cumulative distribution function (CDF) for clients’ observed latencies in a scenario with four shards and 10% cross-shard transactions rate. The aggregated latency shows both single and cross-shard transactions and we can see that, as expected, around 10% of the transactions takes more than 30 seconds to complete. Differently from other sharded systems (e.g., [28]) the protocol does not suffer from a convoy effect [29], that is, cross-shard transactions do not delay single-shard transactions.