Add Tabulator in the Data Grid Performance Comparison

[nclemeur]

I think it would great to see how tabulator is comparing against the other grid libraries. For example this article is comparing bryntum grid to other grid solution.

So I have forked their github repo and added a test for tabulator. You can find it here.

Could anyone from the team review the code in the tabulator folder to make sure I haven't missed anything?

My first results suggests that tabulator is quite performant for the unlocked case, but there is really a big performance hit when using the locked columns module (mostly for the initial rendering which goes from about 50ms to about 2200ms on my computer).

[olifolkerd]

Hey @nclemeur

That is a great idea, i would love to see that in there, i will take a look at the code this weekend and give you a shout.

What were the values from your test?

As for your initial results, if by "locked" you are referring to frozen columns, they yes Tabulator is very inefficient there, especially since the 5.0 rebuild as the frozen columns module hasn't been fully migrated to the new internal event structure yet.

It is part of my plan for the 5.2 release to have a rebuild of that module as there are many inefficiencies that can be improved upon.

Cheers

Oli :)

[nclemeur]

Yes, locked is the equivalent of frozen in tabulator world. Maybe it's worth waiting for 5.2 before asking to integrate tabulator in the tests suite. Anyway, I'll wait for your feedback.

[olifolkerd]

Having had a brief look at your code, there is a couple of efficiency savings that could be done here.

Rather than calling setData in the tableBuilt event, you should pass the data array into the data prop on the table constructor, then the table will be initialized with the data, rather than rendering an empty table then loading the data.

You also then shouldnt need to use the blockRedraw and restoreRedraw function (you should never need to use these for a standard setData call, it will manage this for you, they are more intended for when you are manually doing lots of complex actions)

[nclemeur]

I used the setData so that I have a callback to know when tabulator has finished the updated and then stop the timer. Maybe I can use the data prop and then use the tableBuilt to stop the timer (assuming that's the right event to know when tabulator is done). I'll try that tomorrow.
Thank you for the info about the block/restore functions.

[olifolkerd]

yes, i can see why you did that, which is valid. if you went for the data prop then you would need to listen to the dataProcessed event.

see which approach is faster, i would love to know the results.

[nclemeur]

I did try the approach with the data as a prop and it was slightly faster, but the difference were not huge. By curiosity I ran the tests for ag-grid and extjs-classic on my computer,and got the following results:

So unlocked version is really extremely good, but the locked (frozen) is extremely bad too (at least for the initial rendering)...

[olifolkerd]

Very interesting numbers. Great to hear it outperforms ag grid on the standard render!

Hold off untill after the 5.2 launch and the frozen column score should be much better.

I'm aiming to have 5.2 released by mid April

Cheers

Oli :)

[nclemeur]

For completeness I should probably add a link to the original article

[olifolkerd]

Hey @nclemeur

I have just pushed some updates to the 5.2 branch to address some of the frozen columns efficiency issues.

They will be included in the 5.2 release later this month.

Cheers

Oli :)

[olifolkerd]

Hey @nclemeur

Version 5.2 has just been released. Any chance you could run your benchmarks again and see how we get on?

Cheers

Oli :)

[nclemeur]

Hello,

I have just ran the tests with 5.2 and indeed what an improvment for locked/frozen columns. Please see table below

[olifolkerd]

Awesome news. Would you be ok to add these values to the comparison repo?

[nclemeur]

I have created a PR to integrate into the main repo (see bryntum/grid-performance#3), but I don't have access to the same computer as the one they ran their test suites. So my numbers would not be really comparable. Hopefully they'll do it when they integrate that PR...