Dusten Splan schrieb: Hi Dusten, > One suggestion that I can make is changing your > memory_replacement_policy and cache_replacement_policy to heap LRU. What is your reason for using heap LRU? I've heard/read this before that it performs better than LRU. So it's on my todo-list. But I would like to understand better why heap LRU should be better. > You should graph your iops on your disk so that you can see if you are > having issues there and need more disk. Actually we do graph IO, but only "Bytes Read/Written". I just changed that. Thx, for this hint ;) At peak times we have 500kB/s reads and 250kB/s writes on each of the four disks. I think that the problem is with the trillions of very small files. That is why I thought COSS would bring some improvement here. But first I need to get my around COSS configuration. Cheers, Markus
