Hello,
I feel a bit confused when it comes to the disk I/O figures we are seeing on our Domino servers. Wherever I look in the documentation I read “alwaus put your transaction logs on separate super fast disks”.
Having had disk I/O related issues, I took this to heart and separated data and transaction logs. We did this along with a lot of other things which rendered noticeable performance improvements.
The thing that confuses me is that my 2x15krpm RAID1 disk shows a few precent busy along with negligible iops and data rates. At the same time the data disks can still be stretched to the limit with hundreds to thousands of IOPS.
These are systems with 100-200 simultaneous users running mail and applications.
Could it have something to do with the mode of logging used? In this case it is circular style logging. The servers have loads of RAM and it is quite conceivable that the Linux disk caching has the entire 4 GB of logs in RAM meaning that i/o on the transaction log disk only ever is used for sequential writes.
The reason for wondering is that I am contemplating using SSD drives for my Domino servers and while it is clear that NSF and FT data should benefit greatly from SSD, it seems to me right now that it would be complete overkill to place the transaction logs on separate SSD disks given the IO numbers i see at the moment.
While I am on the subject: What actually passes through the transaction log? Adding, modifying and deleting documents naturally but what about creating/updating views? Anything else?
Thanks,
/Erik