Benchmark Testing

[c-emin]

I ran the same data set on Westfalls BlazorDB vs Magic.IndexDb

~1500 images, reading from Azure and storing into IndexDb

BlazorDB 25 min 55 sec
MagicIndexDb 27 min 5 sec

I was really hoping to see faster times in this area.

[Ard2025]

Could you share both benchmark codes with us?

[c-emin]

I can't share the full code or the exact image content...
But I can share a snippet of the basic code I used to write the images to the database.
As you can see it is straight forward textbook code for adding a record to the Index DB.
The only main difference is that I used NewtonSoft.Json in the Westfall Implementation
and System.Text.Json in Magic IndexedDb.

Westfall BlazorDB

var json = await httpResponse.Content.ReadAsStringAsync();
var dynamicObject = JsonConvert.DeserializeObject<dynamic>(json)!;
JObject jo = JObject.Parse(json);
var payload = jo["payload"];

manager = await _dbFactory.GetDbManager(dbName);

await manager.AddRecordAsync(new StoreRecord<object>()
	{
        StoreName = ProfileImagesTable,
        Record = new
        	{
                Id = item.Id,
                imageFormat = imgfrmt,
                imageBytes = @"data:image/" + imgfrmt + ";base64," + payload["imageBytes"],
                sortOrder = item.SortOrder,
                createdOn = item.CreatedOn,
                modifiedOn = item.ModifiedOn,
                }
        });

Magic Indexed DB

Json_Image payload = (Json_Image)System.Text.Json.JsonSerializer.Deserialize<Json_Class_Image>(await httpResponse.Content.ReadAsStringAsync(), jsonoptions).payload;

profileimagesTableQ = await _MagicDb.Query<ProfileImagesTable>();

await profileimagesTableQ.AddAsync(new ProfileImagesTable
	{
     	Id = item.Id,
	imageFormat = imgfrmt,
     	imageBytes = @"data:image/" + imgfrmt + ";base64," + payload.ImageBytes,
     	sortOrder = item.SortOrder,
     	createdOn = item.CreatedOn,
     	modifiedOn = item.ModifiedOn,
	});

[Ard2025]

I think there is some overhead in the query system. i am going to research it when i get some more time on my hands.

[magiccodingman]

@c-emin Magic So, that's ~4.5% speed difference. This project had much of the internal guts based on the nwestfall project. Because it's a full query engine similar to LINQ, there's for sure going to be a performance difference in Magic IndexedDB versus something that's much closer to effectively a keystore.

Though getting that speed difference down would be preferable. But @Ard2025 if I had to guess. This may be the difference between the Nwestfalls instant no stream interop connection.

Magic IndexedDB has stream capabilities that're auto enabled to prevent any message buffer issues. This delay if re-activated 1500 times is the only thing I can think of that would cause any meaningful slow down. Even the LINQ code shouldn't be an issue. Though there could be some performance issue.

But just as a head up, Newtonsoft is amazing, but there's a reason we converted off that. newtonsoft is significantly slower in almost every capacity.

So the difference if I had to guess would be:
1.) The stream being enabled which does have a slight delay when bulk called that many times. That could in theory cause that 4.5% increase in total time. Or the issue may not be the stream, but how data is custom transferred within this project.

2.) Potentially serialization, but it'd be because of something else, not the library. We do have a full custom serialization process though. Or maybe we have double serialization going on somewhere potentially?

3.) The advanced capabilities built in which differentiates this project from a key store could be effecting the speed, but I'd put this as my lowest guess.

Though, a lot of this also if it's possible for you. Can be avoided if you're able to do more bulk additions instead of re-calling the add method ~1500 times but instead bulk insert with add range. Similar to normal LINQ to SQL tactics. Though, the add method should be pretty darn close. I really think it's my #1 guess. I'll put it on my list to look into as well.

[Ard2025]

@c-emin what are your benchmark results when using AddRangeAsync()