Day: October 7, 2020

If you’re in the beginning stages of your data science credential journey, you’re either about to take (or have taken) a probability class. As part of that class, you’re introduced to several different probability distributions, like the binomial distributiongeometric distribution and uniform distribution. You might be tempted to skip over some elementary topics and just scrape by with a bare pass. Because, let’s face it–the way probability is taught (with dice rolls and cards) is far removed from the glamor of data science. You may be wondering

When am I ever going to calculate the probability of five die rolls in a row in real life?

Click through for the answer and for a chart provides different scenarios for real-world probability distributions.

Sorts use storage (memory and perhaps disk if they spill) so they do have a facility capable of storing rows between loop iterations. In particular, the sorted output can, in principle, be replayed (rewound).

Still, the short answer to the title question, “Do Sorts Rewind?” is:

Yes, but you won’t see it very often.

Deploying the report is seemingly straight-forward, but there are some risks we need to consider:

– What should we do if the report already exists?
– If the dataset exists, what should we do if there are other reports that use this (shared) dataset?

The last item is a bit of an edge-case that we’ll have to dive deeper into, but let’s look at the basic cmdlet first.

Click through to see how, as well as some thoughts on those risks.

When monitoring storage systems, engineers should track a variety of metrics to ensure the systems continue to meet application requirements. Three of the most important and commonly cited metrics are latency, I/O operations per second (IOPS), and throughput. In addition to these three, queue length and I/O splitting can also provide valuable insights into storage performance.

In this article, I discuss all five of these metrics and demonstrate them in action. Despite my focus on these five, they’re not the only important metrics to monitor. For example, engineers should also track storage capacity, device cache usage, controller operations, and storage networks. Even seemingly unrelated components can be a factor, such as low CPU utilization, which can indicate that the processor is waiting on storage to complete requests from the application.

Robert also shows off the new perfmon, which may or may not be better than Perfmon Classic.

I remember when I first started deploying Big Data Clusters, they were on Azure Kubernetes Service utilizing the \$200 credit for first time sign ups. By the time I got around to figuring out how to deploy the BDC, not only was my \$200 credit gone, but I started to incur cost out of pocket.

If only there was a feature that would allow me to stop the VMs in AKS whenever I wasn’t using them. Well, I’m excited to share that Microsoft AKS (Azure Kubernetes Service) came out with a neat feature (currently in preview at the time of the publishing of this post) that allows you to stop and start your AKS cluster by running a simple command. Of course I had to try it out on BDCs and to my surprise it worked. Well, sort of. Let me explain…

In yesterday’s post, I showed how to configure an availability group (AG) to use the RegisterAllProvidersIP=0 when you can’t get clients to connect using the MultiSubnetFailover=true connection string attribute.

I mentioned that you have to make some trade-offs when you set RegisterAllProvidersIP=0, and included this comparison:

But….when if you can eat your cake and have it, too?

In some cases, you’ll have some applications & clients that are not able to use MultiSubnetFailover=true, and other clients that can. Perhaps you’re working on updating a bunch of legacy Java apps to move from old jTDS drivers to the current Microsoft JDBC drivers that properly support MultiSubnetFailover=true. Parts of your codebase have been updated, and you want them to make use of the connection string attribute for fast cross-subnet failover. But other parts of your codebase are still being updated and rely on the RegisterAllProvidersIP cluster parameter to be false. Wouldn’t it be nice to have both?