3 ms·
We had similar thoughts with KeyDB. Particularly about the RAM is scarce bit. We started as an in memory cache but we've been adding in better persistence supp
by jdsully 6y ago
We had similar thoughts with KeyDB. Particularly about the RAM is scarce bit. We started as an in memory cache but we've been adding in better persistence support. It's a lot easier to add persistence to a cache design than to try and speed up a database built under the assumption disks were slow and CPU time didn't matter.
If you're serving data from RAM every cycle counts.
- PeterCorless 6y agoSome people measure "low latencies" in milliseconds. Typical for SSD-based databases. Some in microseconds. Appropriate for RAM-based systems. Others in nanoseconds. I just spoke with an old friend at Cisco working with HFT switches. They're sampling data at the picosecond level. But the smallest time sampling I've personally run across was the 100 femtocecond level at SLAC. So there are utterly different orders of magnitude of scale when people talk about "fast" performance.