Peter, another item which Nina should mention is disparity between server size.
Our helpdesk or infra group was initially thinking about Aware more like an enterprise application farm. And after months of testing in many different Servers, OS's, Layers we have found that it really is not as necessary until your reaching a user threshold per box, or connection.
As support mentioned. The variables to consider are many and at each point could be a cause of latency or concern. However for the most part its "Throughput" from DB to Application to Web Frontend to Internet to Customer to the customers Environment"
When you consider the multitude of typical customer "desktop" issues and there consumer internet services you will find your fighting a possible battle of fine-tuning which is overkill.
So our current principles follow the SaaS standards of APDEX
http://apdex.org/overview.html
We look only at the delivery from our DC to the customer’s location and that latency is resulted in the APDEX score. We use this as our SLA towards the customer and now fine tune based on this.
You need diagnostic software at the DC level and a good understanding of each layer to help troubleshoot and diagnose latency or errors. Following keys like "Low UI or GFX", "Limited View Perspective Queries" and a couple other of our own internal guidelines have allowed us to get a Rich user experience on lower cost cloud equipment.
Note:
We have tested on 512 VD's (Virtual Desktops) to now 16gb mem, 8 core servers (non-virtualized or bare metal)
Scaling:
You can take Aware vertical or horizontal. Meaning when broken down
Tomcat scales horizontally best and with most bang for the buck
JBoss we stick with vertical as Bigger boxes are cheaper these days
MySQL cluster (beign tested now) offers enough Speed and Catastrophe/Safety for High Availability SLA's
Thanks,
Lars
