We have a fairly new implementation of Exchange 2013 CU3 which is already beginning to show performance issues. The design was ratified by both Dell and Microsoft before the production build commenced and we recently had Microsoft review it again as part of the RAP program. Our site spans two data centres and is configured with 3 active\3 passive mailbox nodes and 6 CAS's all of which are VM's running on VMWare 5.5. The mailbox role has 12 CPU's and 96GB RAM, while the CAS's have 4 CPU's and 12GB RAM. The initial architecturewas fleshed out by using the Exchange role calculator and at the time we we overspecced our environment to allow for future growth.
Our main concern is that we are seeing daily increases in the consumption of CPU and memory which eventually causes system instability and users lose access to their mailbox via Outlook RPC connections. This generally takes about a week to occur. It appears that the processes for the MSExchangeRPCProxyAppPool and MSExchangePowershellFrontEndAppPool are the biggest consumers of processing power and I am wondering if there is a way to recycle the app pools without causing system issues, e.g. by putting them into maintenance mode first?
We plan to go to CU6 in about 3 weeks and we aren't in a position to rush this any sooner - and we're not sure that this would assist us anyway.