High Performance CouchDB
At Couchio we are working with a company that makes high performance database appliances. These are complete system stacks where all the components, software and interconnects are carefully chosen, modified and optimized. They can deliver performance and scalability that's often an order of magnitude better than what can be accomplished by assembling your own systems with commodity components, and reducing overall hardware and admin costs.
So if you are a user of Apache CouchDB and are interested in high performance CouchDB servers, we'd love to talk with you to help us shape our plans and offerings, and potentially join an early adopters program. Email me at damien@couch.io.
Posted May 25, 2010 2:15 PM
Comments
in your review of servers have you looked at dragonflybsd ???
arthur ingram, May 25, 2010 11:21 PM
It would be nice to see how these features improve performance
swapcache - This is a mechanism which allows an attached solid state drive to be used to cache data and/or meta-data for other filesystems. The meta-data caching feature is particularly useful for machines which manage millions of files. About 2 million inodes and their related blockmaps can be cached per gigabyte of SSD. DragonFly supports up to 32GB of swap, per swap device (up to four) on i386, and up to 512GB of swap, per swap device, on 64-bit, so a very large number of inodes can be cached. Please see the swapcache(8) manual page for more information.
HAMMER - Our filesystem now supports a REDO log on top of the UNDO log it already had. The filesystem will use the REDO log forward-log write and file-extend operations which allows fsync to accomplish its goals with only REDO log writes. This greatly improves the speed of fsync without weakening HAMMER's fast-recovery feature.
tmpfs - The tmpfs port we did from NetBSD is able to fully utilize the VM page cache on the system as well as swap space, with no duplication of data (unlike MFS) and very good cache characteristics which reduce or eliminate disk activity under nominal operation.
arthur , May 25, 2010 11:22 PM
Post a comment