Performance Concerns: Sqids significantly slower than HashIds in benchmarks

[Pouria7]

Performance Concerns: Sqids significantly slower than HashIds in benchmarks

Hello Sqids team,

I recently conducted a performance benchmark comparing Sqids with HashIds (ullmark/hashids.net), and I wanted to share my findings with you. While I appreciate the new features and improvements Sqids brings over HashIds, I've noticed a significant performance difference that I believe is worth discussing.

Benchmark Results

Here are the results of my benchmark test:

Method	Mean	Error	StdDev	Median	Ratio	RatioSD	Allocated	Alloc Ratio
HashId	4.109 μs	0.0715 μs	0.0634 μs	4.109 μs	1.00	0.00	7.2 KB	1.00
Sqids	54.574 μs	1.4042 μs	4.1182 μs	53.410 μs	12.20	0.50	74.22 KB	10.31

Observations

1. Execution Time: Sqids is approximately 12 times slower than HashIds (mean execution time of 54.574 μs vs 4.109 μs).
2. Memory Allocation: Sqids allocates about 10 times more memory than HashIds (74.22 KB vs 7.2 KB).

These benchmarks were conducted using both encode and decode operations together. However, I observed similar results when testing the encode operation alone.

Questions and Concerns

1. Are you aware of this performance difference?
2. Are there any known reasons for this significant gap in performance?
3. Are there plans to optimize Sqids for better performance in future releases?
4. Do you have any recommendations for users who require performance similar to HashIds while benefiting from Sqids' features?

I appreciate the work you've done on Sqids and I'm looking forward to hearing your thoughts on these performance issues. If you need any additional information or if you'd like me to run more specific benchmarks, please let me know.

[aradalvand]

Hey @Pouria7, thanks for your efforts. Could you please share the actual benchmark code?

[Pouria7]

[MemoryDiagnoser(false)]
public class IdGeneratorBenchmark
{

    private int id = 5732;

    [Benchmark(Baseline = true)]
    public string HashId()
    {
        Hashids hashids = new Hashids();
        string encodedId = hashids.Encode(id);

        var decodedId = hashids.DecodeSingle(encodedId);
        //var decodedIds = hashids.Decode(result);

        return encodedId + decodedId.ToString();
    }

    [Benchmark]
    public string Sqids()
    {
        var sqd = new SqidsEncoder<int>();
        string encodedId = sqd.Encode(id);

        var decodedIds = sqd.Decode(encodedId);

        return encodedId + decodedIds[0].ToString();
    }
}

[Pouria7]

I realized that the initial benchmarks might have been affected by the time taken to create instances of the encoders. To address this, I modified the setup to create instances only once:

[GlobalSetup]
public void Setup()
{
    hashids = new Hashids();
    sqids = new SqidsEncoder<int>();
}

Updated Benchmark Results

After making these changes, here are the new benchmark results:

Method	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
HashId	928.2 ns	3.58 ns	2.99 ns	1.00	0.00	104 B	1.00
Sqids	2,558.4 ns	47.31 ns	44.25 ns	2.75	0.05	3040 B	29.23

1. Execution Time: Sqids is now approximately 2.75 times slower than HashIds (mean execution time of 2,558.4 ns vs 928.2 ns). This is a significant improvement from the previous results but still shows a noticeable performance gap.

2. Memory Allocation: Sqids still allocates significantly more memory than HashIds (3040 B vs 104 B), with an allocation ratio of about 29.23.