posted Dec 25, 2016, 1:42 AM by Sami Lehtinen
updated Dec 25, 2016, 1:43 AM
- Typical cloud service mode is that you'll start with good service and performance. Then gradually worsen the service after trapping customers. That's why vendor lock-in is so important. They can provide absolutely crappy service and still keep charging, because cost of changing vendors / platform is annoyingly high. - This also applies to many other businesses other than cloud services. But in this case it's blatantly true. - Even hardware vendors to this. They firt release model SupaSpeed600. When it receives rave reviews, they'll replace it with model SupaSpeed600(b) which is very slightly cheaper, but actually performs at least 50% worse. This happens all the time. Some times they try to hide the plan B really hard. It might be just some extremely small change in revision or something. Yet customers thinking they'll get the fast model get just screwed. - This is also one of the reasons why vendor lock-in is extremely bad thing. They can just keep cashing and almost blackmailing you, if you don't have any viable options.
- More fun with 'traceability'. Now I'm being asked to remove 'duplicates' from database. That's freaking awesome. But how do we know if something is a duplicate or not if it doesn't have unique consistent tracking id? - It doesnt. -> We can't know for sure if something is a duplicate or not. - That's freaking awesome. - If duplicates aren't removed, there's too many records, but if "duplicates" are removed, it's very highly likely that some data is being detected as duplicate even if it isn't. - How to fix this mess? Well, it's very hard to fix this mess afterwards. Only way to get this kind of stuff would not to mess up the data in the very first place. Best way would be to delete all data and reload it from the trusted source(s) where there are no dupes.
- Studied more Ceph documentation to understand better how it works. Including: RADOS Block Storage / RADOS Block Device (RBD) related configuration and NVMe caching.
- Cloudflare TLS nonce-nse - This article also shows how easy it's fail crypto and how many ways there are implement seemingly similar solution. KW: Initalization Vector, Encryption, Message, Key, Cipher, Nonce, TLS, RC4, CBC, GCM, ChaCha20-Poly1305, BEAST.
- How OVH expand in the USA while preserving its European identity - A very nice article by OVH.
- Checked out Windows Server 2016 feature list. All the Hyper-V, Containers, Virtual TPM, JEA, Shielded VMs, Identity Management stuff isn't really viable. It's great. But personally I don't see use for it right now. Storage Spaces Direct seems to be something which is actually nice for small scale operations. Server containers sound something which could be highly interesting. I've been complaining about Windows Licensing costs with Virtualization, which is usually more expensive than the hardware per month. As well as the huge over head caused by running bloat ware called Windows. Licensing + Overhead make Windows cloud operations highly inefficient. That's why Nano Server image is very interesting. Docker support is also very neat.
- PyPy3 5.5.0 released, but it's not yet plug-in replacement for Python 3. I wish it would be. I don't have real performance issue with Python, so I won't be using PyPy3 as long as it requires 'extra work' to get it working.
- Studied Bitbucket's new features - Pipelines, Large File Storage (LFS), Smart Mirroring and Merge checks. I've been using Bitbucket for my personal projects with great success. Continuous Delivery using Bitbucket Pipelines. The LFS implementation is pretty obvious and very basic, but it works. I've done basically the very same without using LFS. By keeping file hash references in flat files (hex) but not storing the file data itself inside the git repository. For replication purposes and automated replication those could be torrent magnet links, or any content based addressing systems keys. Like IPFS, Freenet, etc, list goes on.