On Thursday 22 March 2007 11:25, Muli Ben-Yehuda wrote: > On Thu, Mar 22, 2007 at 10:55:11AM +0200, Peter wrote: > > I am not an expert on this, but any algorithm that runs in O(1) or > > close to that for the data size you use is a candidate. The data > > size should be obviously less than 2^32 for x86 at least in any > > indexable dimension if you want a reasonably smooth ride. Larger > > things (like 30 million record databases to be used in real-time) > > move up to better things like 64-bit x86 and up. F.ex. 30 million > > records will require less than 7 bits per entry just to keep a > > complete linear index in 3GB of RAM (the maximum usable you can put > > in a x86 32 bit machine). > > With PAE you can stick in a lot more RAM on x86-32, but then your > bottleneck becomes the 3GB of virtual address space for a single > process.
Perhaps some kind of a hardware solution can be used. I.e. attaching a pci with memory and addressing it's 64gb memory either directly (if you have 64 bit bus) or in two phases. It can also be any RAM space size you choose, but it will cost you (each additional bit doubles your costs :) ). -- Regards, Tzahi. -- Tzahi Fadida Blog: http://tzahi.blogsite.org | Home Site: http://tzahi.webhop.info WARNING TO SPAMMERS: see at http://members.lycos.co.uk/my2nis/spamwarning.html ================================================================To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
