Chris Miller, a man who knows how to run the room, presented at IBM ConnectED on cloud considerations when preparing to move to the cloud. Chris took a cloud agnostic approach to deployment. In reality, Chris outlined what you should be asking and thinking as you plan a deployment. Most of the issues come less from server sizing and more from network bandwidth sizing and pricing.
Keywords and definitions
- Business Continuity: Goal is to fail over to “somewhere else” to access everything. More than just a data center.
- Disaster Recovery: not just a physical location, where can I go get my data. Treat as a living breathing site. Can’t just go there without testing once a year.
- Fault Tolerant: Just redundancy.
- Redundancy: an always available path.
- Fully hosted environment
- Get rid of everything I have and put it in the cloud
- Purest form of service with options giving you control levels
- dedicated servers (physical or VM)
- Full management rights to your system
- Part of your existing domain (or not)
- Single SLA
- Backup retention standard with modifications allowed
- Contract bandwidth caps- you get charged a ton of money if you go beyond that.
- and more
Purest form of hosting is simply : Internet –> firewall –> servers –> firewall –> users back in your domain
Issues: network saturation can be a huge bottleneck causing your users to tell you the ‘server’ is slow. You need to ask your network guys on whether you prioritize network usage.
Hybrid Environments
Who runs what, who hosts what, your own domain? A hybrid environment implies that you need things like directory sync servers to ensure up to data user information.
- Some of your servers remain in your own sites
- Components or pieces are hosted by a provider
- Your own domain or theirs
- Data retention may fall to you or the provider, depending on where the server resides
- Bandwidth caps may be in place
- Limited access to servers likely
User Connectivity Options: Best practice is to push users straight to a provider server and NOT routing through your network twice. This can get pretty ugly when you have shared resources like calendaring and room scheduling for “cloud email”
Bandwidth Numbers
This is a look into real customer bandwidth data usage for hybrid and hosted solutions. It’s important because pricing of bandwidth can kill your budget.
What bandwidth can you expect?
- You better know your current utilization
- start analyzing network metering report. You want to know the true saturation of your network. It will tell you if your internal network is ready for traffic to your cloud provider.
- Enable platform statistics if the server gives you the stats. Domino does. Connections doesn’t.
- What is the proposed utilization: e.g. what’s my growth pattern
- Create a model representation of growth and peak
- Then answer the question about whether your network can support this during migration
- Questions to ask
- What is the dedicated bandwidth to your organization
- What is the guaranteed constant and burst
- fyi, statistically, the highest bandwidth usage time is Tuesday just after noon.
- Does the provider have a bandwidth cap monthly?
- Do you get cut off or charged huge fees for exceeding percentile or maximums?
- What is their current network situation
- Ask about their network stats
Chris showed a number of scenarios where he found the 95 percentile of usage. The idea is that most providers charge bandwidth based on percentile of usage and not on your burst. You need to take your mail, hosted applications, and other cloud apps to determine your total bandwidth needs. You also need to take into account seasonal usage that will increase your 95 percentile bandwidth usage.
Firewalls: When you move things to cloud or hosted, you can increase the bandwidth through a firewall or security appliance. This will impact you. If the firewall cannot support the throughput then you have a problem. Firewall now needs to be redundant if it’s not already so.
Don’t forget about migration: Migration will represent sustained bandwidth over many days. That will cost you a lot of money. This impacts you more on cloud email rather than hosted websites without a lot of back-end interactions. Best practice: do migrations like this by copying the data to a drive and doing the sneakernet approach.
