RT @PingCAP: Excited to host Sanjay Mishra, Dir. of Product Mgmt @Google AlloyDB, at #TiDBSCaiLE 2025! ⚡️. 📍 Scaling Real-World GenAI: Less….

My hobby project woes come from my lack of understanding the Folios and the relevant details around it in the Linux kernel. I found a good talk by the author, worth watching:.

Join millions who have switched to Grok.

Finally, I've reached a stage where I'm crashing it all the time due to my incompetence at this :-). Switched to a VM model and it's a lot easier than waiting for 10mins for the server to boot.

The Linux kernel has become quite resilient, working on the Ubuntu kernel v6.8.0 variant. My hacking (figuring out) on a custom driver and kernel module shows all kind of error messages in the logs , page faults, RCU invariant violations etc. I've had to reboot only twice so.

Setting up the build system to write a kernel module or driver has become quite easy with these new LLM tools. The last time I tried this by scouring the Internet it took forever.

RT @PingCAP: 🌍 Can’t make it to TiDB SCaiLE 2025 in person? We’ve got you covered. Registration is NOW OPEN for online attendance! Join u….

ALRs (Almost Local Reads) essentially optimize batching by making the remote coordination cost independent of batch size, the core insight doesn’t seem to be revolutionary AFAIU. Their evaluation demonstrates that the technique works but is limited to specific protocol classes.

A theorem called LAW, not CAP, has been proposed. - Linearizable (linearizability, consistency).- Local Reads (local reads or high-cost reads).- Asynchronous.Until now, only two of these could be satisfied. LS is synchronous replication, RA is Raft, etc. RC seems to correspond to.

IIRC, this paper advocates tiered storage of LSM data, where WAL, L0 and L1 go on the fastest storage and the lower levels on something like S3. Writers need to respect device limits. I think the more important observation is context driven IO. Pass LSM operation type (flush,.

August 25 (Mon) - 26 (Tue) ✨ #WebX ✨🌐 Asia's largest global conference, held again this year 🌈 We look forward to welcoming you at exhibition booth G-94 😀 Please check out #TiDB as a high-performance and highly available data infrastructure in the Web3 domain 🚀.

Does it make sense to write a database focused VM that can support multi-tenancy?.

Really like the SQLite Vtable functionality, was able to map an API to SQL quite easily.

Divide et impera - the strategy for all problems.

The happy path in distributed systems, with no concern for latency and handling errors is fairly straight forward.

Google translate does a decent job of Yasufumi’s blog post talking about some performance tests.

Yasufumi Kinoshita one of the top MySQL engineers left MySQL/InnoDB in May and is working on a fork where he is focusing on performance. I have huge respect for his talent and skills, expecting some solid changes. AFAIK, he is also the original author of Percona’s XtraBackup and.

RT @ngaut1: TiDB at Flipkart: Managing Data at E-Commerce Scale via @YouTube.

Database are sticky, gaining trust is difficult.

"in practice the users try and a consistent load" -> "in practice the users try and maintain a consistent load".

At cursory read it should help in making it dynamic, currently in TiDB the DBA has to do static modeling (well almost) there is a calibrate feature that helps with the allocation of RUs based on historic data because in practice the users try and a consistent load on the DB. I.