Queues, EPYC, BitMaelum, CF/H3, Matrix, Bcache, Search

  1. Today I tested 3 different queue systems: Kafka, ActiveMQ and RabbitMQ. I liked RabbitMQ it's simple to use and got good integration for many frameworks. Also it can be simply protected using username and password. RabbitMQ is just like I generally prefer my software. Quick and simple to install, easy to configure and just works. And lot's of related performance, optimization and architecture discussion. Ref: Advanced Message Queuing Protocol (AMQP) (@ Wikipedia) and Message-oriented_middleware (MOM) (@ Wikipedia).

  2. Replaced older servers with new ones utilizing AMD EPYC (@ Wikipedia) 32-Core CPU's / 64 threads.

  3. BitMaelum (@ GitHub) - Interesting idea of replacing email. Full e2ee, ephemeral dual key exchange, PFS, address portability and so on. Yet they didn't provide solution to the addressing problem. All modern fully decentralized systems use public key hashes ass addresses, like Session and Briar. Then whatever layer is preferred can be used to link names / hashes.

  4. Cloudflare announced that they're pushing HTTP/3 on for more sites using the free plan. But I've had the HTTP/3 option enabled all the time. The HTTP/3 transport protocol had been working well all the time. Nothing changes. kw: HTTP, HTTP3, QUIC, H2, H3.

  5. Got a few burners and installed testing Threema, Signal, Session, Briar and Wickr Me on it. Useful when and if alt com channels are required.

  6. Studied crytocurrency layer-2 vs side-chains. Nice post: Layer-2 for beginners (@ gourmetcrypto.substack.com).

  7. One of major problems of Matrix network is the full mesh topology. Partial mesh with some routing protocol should be utilized to reduce federation traffic between servers. It leaves lot of room for optimization and automatic healing. What if there's something wrong with the links, how many multiple routes being active and so on, when idle connections are closed etc. But that's all technically trivial and comes under topic of optimization. How many hops routing chain can have, and if too many, should the route be just cut short with direct connection to the destination, or some other better (?) distant node? Of course this leads to the question of hot spots in the network and how those are being managed etc. If not managed well, it can lead to strange traffic fluctuations.

  8. Element Android image / media upload retry is broken. It won't work. But by canceling and re-selecting media for upload does work. World is so full of broken state code. I also wish the client would have better encryption key management features as well as option to easily select and delete group of messages / mass delete.

  9. Lidl app on Android doesn't know "Finnish", the language selection in Finland shows German, but all text are still in Finnish. That's funny. Some kind of UI / APP design issue, that they don't have proper language options per region?

  10. Long discussion with other hackers on one platform about my home server. Yes, I'm using Bcache, yes, I'm using SSD caching for my storage drives. No I'm not using dm-cache. I'm using bcache. Yes, it's in writeback mode and the writeback time is set to slightly high value 28800 seconds? But why? Well, I don't want the storage drive to wake (or spin) all the time. But doesn't large transfers wear out the drive quickly? No, that's exactly why there's sequential access cutoff parameter, which doesn't store the files (ahem sequential series of blocks) sized in gigabytes into the cache. The cache itself doesn't know anything about files to be clear, as it's name says, it's b(lock) level cache.

  11. Added my-self to Neeva waiting list as well as tested Right Dao, it's surprisingly good and especially really fast. Thank you for that, and this time it isn't sarcasm. I didn't expect such high performance from a new search engine. With Mojeek latency is at times bit annoying. DDG used to be slow as well, but they've done a good job lately, even as mentioned DDG isn't a (true) search engine, it's meta search engine. Yet DDG masks slow results using annoying technique where page is loaded first and then results. - Update, during posting time - Currently Mojeek is clearly faster than DDG.

  12. I think there's a bug with Google Sites. If I edit page and immediately hit publish the latest changes are lost. I need to first navigate to another page in editor and then hit publish to see the changes. Pressing publish button doesn't seem to trigger the save event like navigating away?