Hi, I'm sure this question has been asked a million times before but I wanted to ask it as I'm unfamiliar with this code base and its history.
For a class I'm taking, I've been reading the paper 'Join Processing in Database Systems with Large Main Memories' by Shapiro which talks about how great hash joins are. I was just wondering if there is a reason why hash joins are not implemented in Drizzle? I would have thought that hash joins might be beneficial in the environments that Drizzle is targeting - those with large amounts of memory. Some of the reasons that I was thinking of for no implementation of hash joins are (feel free to correct me, that is the reason I'm sending this email): - deliberate design choice to leave them out - difficulty of implementing them with the current code case - improvements to query cost from hash joins are not substantial enough to warrant the implementation effort - real queries that yield a large savings from using hash joins over other plans are rare I saw that Mark Callaghan mentioned (http://mysqlha.blogspot.com/2008/10/who-needs-hash-join-intel-ssd-will-soon.html) that hash joins may not be needed due to the price of flash disk falling so maybe this question is irrelevant. Also, from reading the comments on that post, I saw that someone already wrote nearly of the hash join code for MySQL and there is a detailed worklog available on that effort (WL#2241). But apparently, that code can not be reused. Again, I hope the list doesn't mind questions like this. I just think it would be valuable for me to get an opinion from people who actually work on implementing databases instead of just reading about things like this in an academic paper. -Padraig _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
