I would like for the monitoring endpoint configuration to support two new options:

1) How many consecutive times TrafficMgr does not get an http 200 from the endpoint before it deems it down. I have a scenario where I want it to be deemed down immediately after one failure for a quicker failover.
2) How many consecutive times TrafficMgr receives an http 200 after it has deemed it down before it will deem it up again. I have seen a scenario where table storage (or SQL Azure) is inconsistent (fails one query then succeeds the next, back and forth) and I want to tell TrafficMgr not to deem my service up again until it receives 10 consecutive http 200's.

Amazon's DynamoDB offers the ability to scale access to a single table, while Azure Table Storage is limited to 5000 entities/account. With Azure Federations now in place, this seems like the last remaining bottleneck for my application as it grows.

Spreading my data across multiple accounts is a suboptimal solution because (1) it requires an additional layer of complexity in my application, (2) these accounts are not free in that I must request more of them from Microsoft myself every time I need more, (3) I'm not guaranteed to get more than 20 (5 to start, 20 if you go through one round with Azure support), and (4) I have to spend time monitoring and modeling when I should start adding new customer accounts to the next table storage account, which involves the guesswork of estimating future loads of customers.

With the new azure IaaS service going public soon I would really like Microsoft to extend this offering to allow remote desktop as a service or remote application as a service using RDP.

This could be licensed per concurrent connection as a local "External Connector Licensing" counterpart or a per user / device licensing mode.

The trick could be to allow an “RDP gateway vm service role” which deploys multiple boxes in the same "subdomain" and allows me to scale the number of boxes in the role.

There are several scenarios where this would come in handy. For us as an ISV we would use it to deliver our windows software as a SaaS solution to our customers. We are doing this with another provider now. For individual businesses it would allow them to move their software to Azure before upgrading servers / workstations. This would allow them to switch to newer os versions earlier, keeping just old software on the old os. It would also allow businesses to move to a windows 8 native apps only environment because they could move the “old” software to Azure RDP.

IaaS service was announced here:
http://www.zdnet.com/blog/microsoft/microsoft-to-enable-linux-on-its-windows-azure-cloud-in-2012/11508

Currently when using a Performance Policy with Traffic manager it does not intelligently support multiple hosted services within a single data center, instead it sends all traffic to the first hosted service.

I have a scenario where this falls down. We have built a service that is associated with its own storage account and has two scaling flex points. The first is to simply add more compute instances to the hosted service. But eventually it could outgrow the scaling capabilities of a single storage account. So our second scaling flex point is to add more hosted services that each are identical except that they use different storage accounts.

I want to use a performance policy with Traffic Manager because our service is geo-distributed. But if one region gets really hot I want to be able to add more hosted services in a single data center. I think having the Traffic Manager simply use a round robin approach for all hosted services within a single datacenter is a reasonable approach.