What is a Page Split
Tables, and indexes are organized in SQL Server into 8K chunks called pages. If you have rows that are 100 bytes each, you can fit about 80 of those rows into a given page. If you update one of those rows to have more data that contains a variable length field like VARCHAR, NVARCHAR, and others, that will cause the page to overflow forcing a page split. The page split takes about half of the data and moves it into a new page, leaving about half in the original page. Another action that causes a page split is to insert a row that based on the indexing would go into a page that is nearly full, if the inserted row doesn’t fit a page split occurs.
If there is room, and your update or insert doesn’t require a split, this is pretty quick to do since SQL Server is just updating one page and then writing it to disk, and to the transaction log. But if the updated or inserted row doesn’t fit, SQL Server needs to allocate a new page, move about half the rows, and then write both pages to disk and to the transaction log. Additionally the pages in all the indexes that point to the data pages need to be updated. Let’s say your table had 1 clustered index, 4 nonclustered indexes, at a minimum 7 pages would be updated, 1 for the clustered index structure, 4 for the nonclustered indexes, and 2 in the data pages in the clustered index. In this specific example the page split would cause a minimum of 7 times the I/O as an insert or update that didn’t require a page split.