Getting my redundancy right in Azure

posted by Jeff | Tuesday, October 8, 2024, 2:30 PM | comments: 0

Last year I outlined the costs to keep the sites on the air. Things have changed a little for various reasons, and I've learned some things.

First off, the hosted forums are still running, remarkably, on a pair of B1 app service instances, for $24 a month total. That's nuts. Even though the ad revenue tends to be less, the PointBuzz forums run on that platform and tend to have more views. It's surprising because there are only 1.75 gigs of memory. Granted, this is the only thing running on that service. But it's deeply satisfying to see it. The lesser virtual CPU time does mean somewhat slower response times though, averaging around 100ms. That's certainly not bad, but I really like to see sub-50.

The other service has 14 sites running on it, including this blog, some placeholders, my music locker, the non-forum part of PointBuzz and all of CoasterBuzz. It was running on a pair of B2 instances, which includes 3.5 gigs of memory, and it was pretty regularly maxing those out. While not instant, the platform mostly replaced the broken instances, but users would still see some temporary slowness. Sometimes I'd get text alerts too, which is annoying. I did my best to optimize all of the code that streams images from the database, worried that was part of the issue, and did the same on the music locker. It didn't make much of a difference.

What I did notice is that in moments of instability, I'd scale it up to P0V3 instances, and the total memory would just sit completely level and predictable. Odd. I realize that the basic tiers, as described by Microsoft, aren't meant for "production loads," but relatively speaking, the sites aren't particularly heavy loads. P0V3 does have a full 4 gigs of memory, but I don't think half-gig would make that big of a difference. But still, the data was indisputable in terms of stability. See the graph below. Each line is memory usage by an instance. So erratic! Then, mid-Monday, you can see it levels off, and the two instances sit right in the 83% zone and don't move.

I don't know the reason for this. I would assume that something is configured differently in the basic instances compared to the newer v3 "premium" tier. Sure, they turn some features off, but I would think the basic container running infrastructure would be more or less the same. Again, the hosted forums have a similarly smooth usage pattern.

I sucked it up and subscribed for a year on the premium, which will cost about $16 more a month. It's not like I'm making money anyway, so might as well just be happy with the performance.


Comments

No comments yet.


Post your comment: