Key notes from the "scaling up and scaling out" session:
Cost: scale-up and scale-out come out at cost, however while talking about high availability the focus should not be on cost
what's the best solution: There are no "perfect" solutions. Different application can use different strategies to achive scale-up/scale-out/high availability solutions
Memory: Memory is king.. use it wisely. Look into key_reads/key_read_request miss rates (not just key_buffer_size)..
Disks: Use RAID if you are no doing so. Use the right interface/disk. Consider CRAM or CPU load
Partitioning: I agree... spread the load. Key benefits of partitioning in general: spread the information, smaller data/index gfragments, parallelizations (better DML). examples can be logging data (for example based on date). partition is daily, weekly or monthly basis. usually applications are looking for past x days of information . By Paritioning on date key you will end up accessing smaller set of data/index segments (plus dropping older data is so easy!!)
replication:spread reads..
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment