... journaling will add some I/O workload to your system, for each update operation the data has to be written two times, but it's treated as redundant data and only one of both is synchronous I/O (the record or the journal, depends on commit or not). The performance impact could be minimized by caching, or by providing a dedicated ASP for the Journalreceiver, so that there is no repositioning of disc arms for this sequential write operations.
The main factor is IMHO the balance of the system, if you have enough disc arms, there will be nearly no performance degradation (and 5 % degradation is not really measurable). One of the problems might be some boxes out with nearly as much CPUs as disc arms and heavy usage of unprotected storage (= swap for missing main memory) - her you will see maybe the numbers Charles was writing of.


This thread ...

Return to Archive home page | Return to MIDRANGE.COM home page