Secondly, what is the cost of your downtime to the business either in a monetary amount or customer satisfaction. Not everyone knows this but there is always a value or “what is the consequence to your business if the service is unavailable?”

By evaluating these three areas you can come up with a TCO and you will be surprised at how low the break-even point is when it is more cost effective to use ft technology.

Here is the challenge – send me the above information an I will work it out for you.

I should point out also that all the redundant components in the world won’t protect you from some disasters. You can have a beautifully designed, redundant system…but if there is only one of any component, there is a single point of failure. (Example: the motherboard.) When you dream dreams of high availability you don’t dream dreams of RAIDed disks or ECC RAM. You RAID at a system level, or even at a rack level.

Clustering, hadoop, openstack…this is the start. Truly redundant systems have redundant networks, storage, virtual servers…you name it! They will even have a fibrelink to an off site location and real-time block-level replication on the storage systems such that disaster recovery time is measured in seconds.

All of the above said, you can still do an absolutely amazing level of redundancy with your primitive off-the-shelf whitebox servers and some mental elbow grease. When you are an SMB sysadmin like me, “high availability” simply has a different meaning than it does to those whose resources are orders of magnitude larger.

What a large enterprise spends on the management software for a single project whilst chasing the “continual availability dream” could fund my IT spend – including salaries – for a decade. The funny part? My staff and customers are just as happy with “less than 10 cumulative hours unscheduled outages per year” as they are with the “continuous availability” of large enterprises like banks.

People understand the need for maintenance. Schedual it properly and with a little bit of customer expectation management you can turn a percieved need for “continual availability” into a satisfactory and achievable “high availability.” On budget, and without burning our your sysadmins.

Regards,
Jodi E.
Microsoft SMB Outreach Team
msftoft@microsoft.com

One more point that I think may matter: it’s not just that they’re limited to a single core, performance is severely limited in both the single core VM as well as the overhead taken out of the system. Their method, check-pointing, consumes a lot of CPU and IO resources to operate, meaning that VMWare’s software FT contains a huge performance cost to not only the fault tolerant VM, but leaves far less CPU cycles and IO bandwidth for the remaining non-FT VMs.

Disclaimer: I’m a hardware engineer for Stratus Technologies.

The systems are primarily architected around the concept of an ultra reliable host. I would be delighted to work with you and if you need specific commercial, please do contact me at andy.bailey@stratus.com if I can be of assistance. I am not sure if you have picked up on our availability guarantee – we are now offering Zero Downtime for VMware. So you can virtualize your Tier one applications with no downtime, no failure and no fear. We will give you $50K if you have any downtime. You also raise the interesting question of dr – couple the solutions with other interesting technologies from the virtualisation and storage vendors and you get an environment that is the best in the industry. But first, I need to understand exactly what you need to do. Andy.

I’ve just run into Stratus and your blog although I’ve been doing a fair bit of work with clients on high scalability and availability solutions. I’m interested in hearing more about the cost difference (say per server) you’re talking about here (as my clients ask me about this all the time).

Also, do you have solutions for disaster recovery as well that avoid/minimize data loss beyond what is common in the industry? If so, do you have numbers characteristic for various industries (finance, online gambling, etc) ?

Interested in hearing more.

Thanks,
– Udi Dahan

I’m just now getting a little time to catch up on your blog. I read your review of my article with great interest.

Your points are quite right. However, nothing is perfect. The problem that active/active solves is the failure of a site. If a site fails, and there is no backup, all is lost (the press is full of such stories – check out my Never Again articles in the Availability Digest). Active/active provides two or more geographically-separated sites so that you keep going no matter what. True, there is some data loss if asynchronous replication is used – typically tens to hundreds of milliseconds, but at least you are up-and-running. A solution using just co-located fault-tolerant systems is not disaster tolerant. A site disaster will take down the system. The best you can do is to replicate to a remote site. Assuming that you are using asynchronous replication, you will also lose some data. And recovery could be minutes or hours rather than seconds or subseconds as is being achieved with active/active systems.

By the way, if you use synchronous replication, no data is lost and there is no split-brain mode. You are limited to the disatance between nodes by performance concerns (typically tens of kilometers), but that is better than having all of your eggs in one room.Of course, you’ve got to be pretty clever on how you remove the failed system from the scope of further transactions so that processing can proceed, but there are several solutions to this problem.

Why not add an active/active option to ftServer to make it disaster tolerant and fully faiult tolerant?

- Bill