Let's say you're making a site like reddit and you want to have infinitely scrolling posts (25 at a time, and 25 more loaded when you reach the bottom)
The naive solution uses LIMIT AND OFFSET, but these are not entirely desirable due to performance considerations and also because if a new post is added to a page while you're on it, when you navigate to the next page, there will be a duplicate.
So solutions instead recommend using a WHERE clause and then sorting by some discrete, unique value. When loading more pages you pass the current value and then the database knows where to go from there.
I would use that, except in my situation these values are not unique. For example, one of the sorting options for posts involves non-unique values. So the former solution wouldn't work because it would possibly contain duplicates.
One idea I had is to just return 1,000 or so post id's on the initial page load. Then the client would send the first 25 post ids to the server to retrieve those. If they scrolled down to the bottom then they would send the next 25 to get the post data for those, etc.
The only issue with this is that it is also not perfectly efficient. If the user doesn't scroll at all it was a waste to send those 1000 post ids.
Is there a proper solution to this? How does one efficiently handle pagination that eliminates duplicates when the sorting option involves non-unique values?
Just add an unique column as a last column to sort on. If the user wants to sort his books by "author, title", just change it to "author, title, book_id".
I currently make some async ajax calls and creates rows in table based on returning data. If rows are around 400-500, page hangs after this dom creations. eg. if i click any text box or drop-downs, it stuck forever. If rows are around 100-200 and then i click any text box, it is still slow but at least it does not stuck.
So, I think the problem is there are too many dom to be created and this causes some problems in browser or page or whatever.
Do you have any ideas or any solutions to improve this performance?
You need to lazy load your data somehow. Ever noticed on sites like Twitter, Facebook, and others that when you scroll to the bottom of the page it will begin loading more records from the server? Good apps will start to garbage collect old records that have been scrolled up as well.
When you scroll through your Facebook news feed it's not loading all your friends post since 2007 into the browser all at the same time. Once a maximum number of posts exists in the DOM Facebook will start removing the oldest ones you scrolled up to make room for more and grab fresh posts from the server so you can continue scrolling. You can even see your browser scroll bar jump up as you scroll down because more posts are being added to the DOM.
No browser is going to be able to handle that much data. You're going to have to sit down and think of a better way to show that data to the user. Only you will know what experience your users will actually want, but no matter what you'll definitely have to reduce the amount of elements you're including on the page.
Example:
Notice how the browser scroll bar jumps up a bit when it gets to the bottom. Twitter gets to the bottom and then loads more data to scroll through. It will eventually start cleaning up data at the top of the page as well if I scroll far enough.
The simplest solution is probably going to be for you to pass up a page number with your ajax requests and have your server only return the results for that page of data.
I want to show a listbox(Table) with nearly 20 Million rows.
How can I do so, with lower memory usage and not letting my server die(stop responding) while doing so.
Even you have any Theoretical idea please do share(I will try to implement).
Need solution very urgently.
I know I cannot load all the rows at once. I need to ask new rows from server every time I scroll. I have tried it but my scroll is not smooth enough.
Thanks & Regards,
Aman
Why not just retrieve the first 100 entries and then once the client scrolls to the bottom you append another 100 entries and so on.
Maybe you could wait for ZK's new feature.
Reference
http://books.zkoss.org/wiki/Small_Talks/2012/March/Handling_a_Trillion_Data_Using_ZK
You could use http://books.zkoss.org/wiki/ZK Developer's Reference/MVC/View/Renderer/Listbox Renderer.
public void render(Listitem listitem, Object data, int index)
To start, you can implement render in way so that you get element to render from datasource at hand by index from render method. You can use standard cache (if Hibernate is in place) or custom-written one if otherwise (look also at EhCache).
#Erik solution is really fast to implement. To add you could make a button, so that user would be aware that loading more records would cost some time and would think if one really needs to load more. Scrolling can make you Listbox just hang up for a moment.
And always make an upper constraint on maximal number of records you will show at one time - don't pollute your server's memory.
Make paging for the table values and retrieve specific number of records on demand.
Use can use dataTable pluging to make pagination for data records.
Notice that you can retrieve data in synchronous and asynchronous way using this library
I am working on some custom jQuery/javascript navigation for a site and I am curious about the performance implications of a design decision.
The way it works is for every option there are up to 8 child options, this hierarchy can go 4 levels deep. I believe this makes for 8^4 or 4096 possible navigation items (probably less but this is the max). These relationships are defined on the server side.
Currently I am working with test data, so there are only about 50 navigation items. When the page loads, I create every navigation item and then only display what is needed for the current selection.
Should I consider rewriting this to only load the items that are needed when a selection is made via an AJAX call or something? I am concerned that my current approach may not scale well if it goes up to 4096 navigation items.
If having 4096 navigation items is a real possibility then you'll have to do something like what you're describing. Simply loading the items into the DOM will take considerable time and further processing will cause greater delays and a poor experience.
For a small number of items, it probably isn't worth your while to over-engineer the solution. However, the performance gains on a large number of items would be expected to be significant.
Here is an example of on-demand loading in a Telerik Treeview. I'm not advocating purchasing the controls (great controls but expensive) however it is an excellent example of what is possible. Coding this on your own wouldn't be difficult to do and, as you can see, makes for a great user experience.
My two cents: if you have the time, do it now before things get even more complicated/difficult to do later.
Downloading them all at the same time is definitely an option, though loading them into the DOM is another story. If you really reach the 4096 possibility limit, you can be looking at pushing down 1-2 megabytes a page load (not to much considering image sizes). Unless you are looking at more data (maybe 16 nodes, 8 levels deep 16^8), then it would be a valid concern.
you could always load 2 deep (8^2 = 64), then when they open a panel, load everything for that panel. The second layer they need to click through should give you enough time to load the rest of the values.
Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 8 years ago.
Improve this question
I need to present a large number of rows of data (ie. millions of rows) to the user in a grid using JavaScript.
The user shouldn't see pages or view only finite amounts of data at a time.
Rather, it should appear that all of the data are available.
Instead of downloading the data all at once, small chunks are downloaded as the user comes to them (ie. by scrolling through the grid).
The rows will not be edited through this front end, so read-only grids are acceptable.
What data grids, written in JavaScript, exist for this kind of seamless paging?
(Disclaimer: I am the author of SlickGrid)
UPDATE
This has now been implemented in SlickGrid.
Please see http://github.com/mleibman/SlickGrid/issues#issue/22 for an ongoing discussion on making SlickGrid work with larger numbers of rows.
The problem is that SlickGrid does not virtualize the scrollbar itself - the scrollable area's height is set to the total height of all the rows. The rows are still being added and removed as the user is scrolling, but the scrolling itself is done by the browser. That allows it to be very fast yet smooth (onscroll events are notoriously slow). The caveat is that there are bugs/limits in the browsers' CSS engines that limit the potential height of an element. For IE, that happens to be 0x123456 or 1193046 pixels. For other browsers it is higher.
There is an experimental workaround in the "largenum-fix" branch that raises that limit significantly by populating the scrollable area with "pages" set to 1M pixels height and then using relative positioning within those pages. Since the height limit in the CSS engine seems to be different and significantly lower than in the actual layout engine, this gives us a much higher upper limit.
I am still looking for a way to get to unlimited number of rows without giving up the performance edge that SlickGrid currently holds over other implementations.
Rudiger, can you elaborate on how you solved this?
https://github.com/mleibman/SlickGrid/wiki
"SlickGrid utilizes virtual rendering to enable you to easily work with hundreds of thousands of items without any drop in performance. In fact, there is no difference in performance between working with a grid with 10 rows versus a 100’000 rows."
Some highlights:
Adaptive virtual scrolling (handle hundreds of thousands of rows)
Extremely fast rendering speed
Background post-rendering for richer cells
Configurable & customizable
Full keyboard navigation
Column resize/reorder/show/hide
Column autosizing & force-fit
Pluggable cell formatters & editors
Support for editing and creating new rows."
by mleibman
It's free (MIT license).
It uses jQuery.
The best Grids in my opinion are below:
Flexigrid: http://flexigrid.info/
jQuery Grid: http://www.trirand.com/blog/
jqGridView: http://plugins.jquery.com/project/jqGridView
jqxGrid: https://www.jqwidgets.com/
Ingrid: http://reconstrukt.com/ingrid/
SlickGrid http://github.com/mleibman/SlickGrid
DataTables http://www.datatables.net/index
ShieldUI http://demos.shieldui.com/web/grid-virtualization/performance-1mil-rows
Smart.Grid https://www.htmlelements.com/demos/grid/overview/
My best 3 options are jqGrid, jqxGrid and DataTables. They can work with thousands of rows and support virtualization.
I don't mean to start a flame war, but assuming your researchers are human, you don't know them as well as you think. Just because they have petabytes of data doesn't make them capable of viewing even millions of records in any meaningful way. They might say they want to see millions of records, but that's just silly. Have your smartest researchers do some basic math: Assume they spend 1 second viewing each record. At that rate, it will take 1000000 seconds, which works out to more than six weeks (of 40 hour work-weeks with no breaks for food or lavatory).
Do they (or you) seriously think one person (the one looking at the grid) can muster that kind of concentration? Are they really getting much done in that 1 second, or are they (more likely) filtering out the stuff the don't want? I suspect that after viewing a "reasonably-sized" subset, they could describe a filter to you that would automatically filter out those records.
As paxdiablo and Sleeper Smith and Lasse V Karlsen also implied, you (and they) have not thought through the requirements. On the up side, now that you've found SlickGrid, I'm sure the need for those filters became immediately obvious.
I can say with pretty good certainty that you seriously do not need to show millions of rows of data to the user.
There is no user in the world that will be able to comprehend or manage that data set so even if you technically manage to pull it off, you won't solve any known problem for that user.
Instead I would focus on why the user wants to see the data. The user does not want to see the data just to see the data, there is usually a question being asked. If you focus on answering those questions instead, then you would be much closer to something that solves an actual problem.
I recommend the Ext JS Grid with the Buffered View feature.
http://www.extjs.com/deploy/dev/examples/grid/buffer.html
(Disclaimer: I am the author of w2ui)
I have recently written an article on how to implement JavaScript grid with 1 million records (http://w2ui.com/web/blog/7/JavaScript-Grid-with-One-Million-Records). I discovered that ultimately there are 3 restrictions that prevent from taking it highter:
Height of the div has a limit (can be overcome by virtual scrolling)
Operations such as sort and search start being slow after 1 million records or so
RAM is limited because data is stored in JavaScript array
I have tested the grid with 1 million records (except IE) and it performs well. See article for demos and examples.
dojox.grid.DataGrid offers a JS abstraction for data so you can hook it up to various backends with provided dojo.data stores or write your own. You'll obviously need one that supports random access for this many records. DataGrid also provides full accessibility.
Edit so here's a link to Matthew Russell's article that should provide the example you need, viewing millions of records with dojox.grid. Note that it uses the old version of the grid, but the concepts are the same, there were just some incompatible API improvements.
Oh, and it's totally free open source.
I used jQuery Grid Plugin, it was nice.
Demos
Here are a couple of optimizations you can apply you speed up things. Just thinking out loud.
Since the number of rows can be in the millions, you will want a caching system just for the JSON data from the server. I can't imagine anybody wanting to download all X million items, but if they did, it would be a problem. This little test on Chrome for an array on 20M+ integers crashes on my machine constantly.
var data = [];
for(var i = 0; i < 20000000; i++) {
data.push(i);
}
console.log(data.length);
You could use LRU or some other caching algorithm and have an upper bound on how much data you're willing to cache.
For the table cells themselves, I think constructing/destroying DOM nodes can be expensive. Instead, you could just pre-define X number of cells, and whenever the user scrolls to a new position, inject the JSON data into these cells. The scrollbar would virtually have no direct relationship to how much space (height) is required to represent the entire dataset. You could arbitrarily set the table container's height, say 5000px, and map that to the total number of rows. For example, if the containers height is 5000px and there are a total of 10M rows, then the starting row ≈ (scroll.top/5000) * 10M where scroll.top represents the scroll distance from the top of the container. Small demo here.
To detect when to request more data, ideally an object should act as a mediator that listens to scroll events. This object keeps track of how fast the user is scrolling, and when it looks like the user is slowing down or has completely stopped, makes a data request for the corresponding rows. Retrieving data in this fashion means your data is going to be fragmented, so the cache should be designed with that in mind.
Also the browser limits on maximum outgoing connections can play an important part. A user may scroll to a certain position which will fire an AJAX request, but before that finishes the user can scroll to some other portion. If the server is not responsive enough the requests would get queued up and the application will look unresponsive. You could use a request manager through which all requests are routed, and it can cancel pending requests to make space.
I know it's an old question but still.. There is also dhtmlxGrid that can handle millions of rows. There is a demo with 50,000 rows but the number of rows that can be loaded/processed in grid is unlimited.
Disclaimer: I'm from DHTMLX team.
I suggest you read this
http://www.sitepen.com/blog/2008/11/21/effective-use-of-jsonreststore-referencing-lazy-loading-and-more/
Disclaimer: i heavily use YUI DataTable without no headache for a long time. It is powerful and stable. For your needs, you can use a ScrollingDataTable wich suports
x-scrolling
y-scrolling
xy-scrolling
A powerful Event mechanism
For what you need, i think you want is a tableScrollEvent. Its API says
Fired when a fixed scrolling DataTable has a scroll.
As each DataTable uses a DataSource, you can monitoring its data through tableScrollEvent along with render loop size in order to populate your ScrollingDataTable according to your needs.
Render loop size says
In cases where your DataTable needs to display the entirety of a very large set of data, the renderLoopSize config can help manage browser DOM rendering so that the UI thread does not get locked up on very large tables. Any value greater than 0 will cause the DOM rendering to be executed in setTimeout() chains that render the specified number of rows in each loop. The ideal value should be determined per implementation since there are no hard and fast rules, only general guidelines:
By default renderLoopSize is 0, so all rows are rendered in a single loop. A renderLoopSize > 0 adds overhead so use thoughtfully.
If your set of data is large enough (number of rows X number of Columns X formatting complexity) that users experience latency in the visual rendering and/or it causes the script to hang, consider setting a renderLoopSize.
A renderLoopSize under 50 probably isn't worth it. A renderLoopSize > 100 is probably better.
A data set is probably not considered large enough unless it has hundreds and hundreds of rows.
Having a renderLoopSize > 0 and < total rows does cause the table to be rendered in one loop (same as renderLoopSize = 0) but it also triggers functionality such as post-render row striping to be handled from a separate setTimeout thread.
For instance
// Render 100 rows per loop
var dt = new YAHOO.widget.DataTable(<WHICH_DIV_WILL_STORE_YOUR_DATATABLE>, <HOW YOUR_TABLE_IS STRUCTURED>, <WHERE_DOES_THE_DATA_COME_FROM>, {
renderLoopSize:100
});
<WHERE_DOES_THE_DATA_COME_FROM> is just a single DataSource. It can be a JSON, JSFunction, XML and even a single HTML element
Here you can see a Simple tutorial, provided by me. Be aware no other DATA_TABLE pluglin supports single and dual click at the same time. YUI DataTable allows you. And more, you can use it even with JQuery without no headache
Some examples, you can see
List item
Feel free to question about anything else you want about YUI DataTable.
regards,
I kind of fail to see the point, for jqGrid you can use the virtual scrolling functionality:
http://www.trirand.net/aspnetmvc/grid/performancevirtualscrolling
but then again, millions of rows with filtering can be done:
http://www.trirand.net/aspnetmvc/grid/performancelinq
I really fail to see the point of "as if there are no pages" though, I mean... there is no way to display 1,000,000 rows at once in the browser - this is 10MB of HTML raw, I kind of fail to see why users would not want to see the pages.
Anyway...
best approach i could think of is by loading the chunk of data in json format for every scroll or some limit before the scrolling ends. json can be easily converted to objects and hence table rows can be constructed easily unobtrusively
I would highly recommend Open rico.
It is difficult to implement in the the beginning, but once you grab it you will never look back.
I know this question is a few years old, but jqgrid now supports virtual scrolling:
http://www.trirand.com/blog/phpjqgrid/examples/paging/scrollbar/default.php
but with pagination disabled
I suggest sigma grid, sigma grid has embed paging features which could support millions of rows. And also, you may need a remote paging to do it.
see the demo
http://www.sigmawidgets.com/products/sigma_grid2/demos/example_master_details.html
Take a look at dGrid:
https://dgrid.io/
I agree that users will NEVER, EVER need to view millions of rows of data all at once, but dGrid can display them quickly (a screenful at a time).
Don't boil the ocean to make a cup of tea.