CPU load, Crypto Wars, Data Protection, Skype, Deepmind, FLIF, ServerBear, Tempfile
Post date: May 12, 2016 4:50:48 PM
- CPU load measuring fails. I complained about one process being sucky and using way too many CPU cycles, working inefficiently and providing very bad response times for messages arriving via one pub/sub queue system. Guess what was the developers answer? I've been checking it out and it seems to only use about 13% of CPU why is that a problem. - Thank you for that. Btw. My guess is that the app is single threaded and the developer used 8 core CPU. I've seen fail comments like this with Anti-virus tools evaluations too. Anti-virus X is great, because it uses only 25% or 50% of CPU but Anti-virus software Y caused slowness because it used 98-100%. Excellent job guys! How about writing an article 'what went wrong' about your own article? CPU percentage can be shown differently, other tools show load on on whole system and other show / core. Which means that program which utilizes threading efficiently can have much more than 100% going on.
- All this discussion about privacy and 'crypto wars' reminded me that there are other fundamental rights and regulations which dictate what can and what must be done. Here's a free handbook for European Data Protection. I've read it, it's at the same time horribly boring and yet very interesting at times. I guess most of people don't know enough about data regulation, because so many laws and rules are seriously conflicting on that sector. I guess in many cases it's just best to try what's sensible and can be easily and sensibly argued. But knowing if it's actually legal is going to be extremely hard and nobody knows that before extensive court case. Ouch.
- Skype direct connections only? Does anyone get this? I don't get it. If direct connections are the only way to connect, then I would assume that cloud wouldn't be used, and IP would be always revealed. If the statement would be disallow direct connections or always use indirect connections, then I would understand that the IP address won't get revealed. Am I missing something? Quite from UI: "When you enable direct connections only, your IP address will be kept hidden when you cal people who aren't on your people list. This may delay your call setup time."
- Google's Deepmind - A nice article about Artificial General Intelligence (AGI), AI. Yes, it's very shallow not deep at all. But nice on general level because it talks about General AI. Ha.
- I've posted about FLIF image format earlier, but here's my latest comments: " FLIF provides excellent lossless image compression. The best part is that it's free and not patent encumbered. Yet in some cases lossy compression would be still more usable. But it's important to have options. I personally expected that JPEG2000 would have made a mark, but nobody seemed to care. Of course todays extensive image use on web browsing might make a difference. Also CPU's are faster so algorithms which require more CPU power to decompress are more usable than ever before and I guess that pretty much stands for future too. I remember time when GIF decompression was slow and JPEG decompression was ridiculously slow."
- I did run ServerBear tests for my Scaleway server (Type VC1 x86 VPS). Results are here.
- Reminded myself about tempfile. Which is very useful when working with data which is in 'log form' and can be processed in blocks rows and doesn't fit in system RAM. In such case performance is much better than using database for this kind of 'batch job' or temporary storage. The SpooledTemporaryFile is awesome, because it works like /tmp (tmpfs) on Linux and data is kept in RAM if there's no memory pressure, but can be written out to disk, when required. This is optimal for cases where the same program is run on wildly varying environments with large RAM & data set size differences. Some servers got large data but low RAM and in some cases it's just other way around. This is the optimal generic solution.
- Something different? Quickly checked out Mi-28.