News for hackers. Stuff that matters.
Re:Caching Beyond RAM: The Case for NVMe by mmt on Friday June 15, 2018 @12:52AM (#17317610)
What would be interesting to see is if this benefit would be better applied directly to the database instead.

The relative scarcity of NVMe ports/bandwidth per server may make that as unattractive as doing the (RAM) caching on the database server itself, but it's not obvious, if one could only spend the money in one place, where it would be best spent.

Re:Caching Beyond RAM: The Case for NVMe by rzzzt on Friday June 15, 2018 @02:32AM (#17317887)
NVMe drives use 2-4 PCIe lanes, so you can have quite a few of them in a system with a suitable adapter.
Re:Caching Beyond RAM: The Case for NVMe by mmt on Friday June 15, 2018 @02:23PM (#17322183)
I'd say that's comparable scarcity to DIMM slots, as the ratio between those and PCIe channels tends to be in the range of 1:2-4 on current CPUs.
Re:Caching Beyond RAM: The Case for NVMe by dragontamer on Friday June 15, 2018 @10:21AM (#17319947)
Intel seems to be all in on NVDIMMs. IE: Adding more "RAM" slots where Optane can fit into.
Re:Caching Beyond RAM: The Case for NVMe by mmt on Friday June 15, 2018 @02:33PM (#17322281)
I would expect that to favor memcache, until the RDBMSes get code to specifically take advantage of it, but even that may be a naive expectation.
Re:Caching Beyond RAM: The Case for NVMe by ihsw2 on Friday June 15, 2018 @08:32AM (#17319260)
AMD's SP3 (Epyc) offerings address the limited capacity. Compared to Intel's current offerings, they are much more cost-competitive in terms of density.
Re:Caching Beyond RAM: The Case for NVMe by mmt on Friday June 15, 2018 @03:01PM (#17322542)
That's certainly true for PCIe lanes, though their memory channel density is unremarkable. (I've also read about cache latency affecting database performance, but I'm not sure what to think of that consindering how synthetic benchmarks usually are).

Historically, I've had trouble convincing management to get anywhere close to reaching the capacity of a server's PCIe lanes, possibly because of a general ignorance around direct attached (but external) storage options, but the popularity of NVMe may change that.

Now the challenge may become convincing people that 128 lanes per CPU is better than 40.

There's also the (extension of the perpetual challenge) that the mere existence of such a price-competitive option to Intel means one doesn't have to panic at the slightest hint of scale and invest immense engineering effort and non-linear infastructure cost into a distributed database to replace a "single"/central [1] RDMBS.

[1] With all the usual replicas and caches for performance (and failover), as is the point of the article.