TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

How to Build a Non-Volatile Memory DBMS

96 pointsby blopeurabout 8 years ago

8 comments

hyc_symasabout 8 years ago
Eh. None of this is new, and we already anticipated this with LMDB back in 2009.<p>The fact that NVRAM is directly addressable (and thus can bypass the page cache) will eventually play out as irrelevant. It will always be a fact that slower mass-storage will exist, more cheaply than fast in-core storage, persistent or not. The page cache will still be needed even for NVRAM, and the &quot;page&quot; will still be the necessary atomic unit of memory interchange. (Direct access is of course a great thing, but it will be direct access to <i>virtual</i> addresses. Virtual memory, and paging in&#x2F;out between primary and secondary storage, is never going away. Every commodity system that has tried to do without PMMUs has failed, for numerous good reasons.)<p>The continual allusions to &quot;frequent writes can destroy memory cells&quot; seems to mainly relate to the extremely short lifetime of Intel&#x27;s 3DXpoint memory, which is from every measure a total failure.<p><a href="http:&#x2F;&#x2F;semiaccurate.com&#x2F;2017&#x2F;03&#x2F;10&#x2F;intel-mislead-press-xpoint-next-week&#x2F;" rel="nofollow">http:&#x2F;&#x2F;semiaccurate.com&#x2F;2017&#x2F;03&#x2F;10&#x2F;intel-mislead-press-xpoin...</a><p>It would be best to ignore 3DXpoint and just focus on STT-MRAM, which is already at parity with DRAM for endurance. (But still lacking in density.)<p>Much of the other stuff in those slides is still off the mark. E.g., LMDB today has perfect crash reliability with zero recovery time. The &quot;write behind logging&quot; they propose still has logging overhead and non-zero crash recovery time - which equals wasted work. Anything that requires logging or any form of compaction or garbage collection is wasted work. It&#x27;s completely unnecessary, and that has nothing to do with NVM. Treating NVM DB design as if it&#x27;s an entirely new and different animal is frankly ignorant. The right design works in all scenarios - as LMDB does.
评论 #14395428 未加载
评论 #14451811 未加载
评论 #14396287 未加载
nuopnuabout 8 years ago
The presentation and the paper pretty much describe how LMDB works already.<p><a href="https:&#x2F;&#x2F;symas.com&#x2F;lightning-memory-mapped-database&#x2F;" rel="nofollow">https:&#x2F;&#x2F;symas.com&#x2F;lightning-memory-mapped-database&#x2F;</a><p>Except, it allows only a single writer and it&#x27;s not directly byte-addressable, unless you make a request to preallocate a chunk of fixed-size mmap&#x27;ed memory.
评论 #14451838 未加载
评论 #14393682 未加载
0xFFCabout 8 years ago
I am currently thinking about stating graduate school and this dude &quot;Andy Pavlo&quot; is my hero, the amount of information he pumps (I don&#x27;t know any other world) to his listeners is outlandish while staying fun and serious and practical at the same time. I watched most of his lectures. I have done some work in industrial reaseach lab with cuda. I did OS a lot and it was my first choice. But after watching his lectures i am seriously considering changing my field to DB. Specially i can do use my Operation Research knowledge in DB much more effective than OS.
评论 #14394414 未加载
crispyambulanceabout 8 years ago
&quot;Non Volatile Memory&quot; really means Intel&#x27;s &quot;3D X-point&quot; memory-- estimated availability of this product for system memory is 2018-ish.<p><a href="http:&#x2F;&#x2F;www.tomshardware.com&#x2F;news&#x2F;intel-optane-cascade-lake-dimm,34471.html" rel="nofollow">http:&#x2F;&#x2F;www.tomshardware.com&#x2F;news&#x2F;intel-optane-cascade-lake-d...</a>
评论 #14393715 未加载
评论 #14393345 未加载
评论 #14393090 未加载
michaelmiorabout 8 years ago
Video is available <a href="https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=ljrpXVlkQ84" rel="nofollow">https:&#x2F;&#x2F;www.youtube.com&#x2F;watch?v=ljrpXVlkQ84</a>
评论 #14393995 未加载
gghhabout 8 years ago
anyone knows if a preprint of the paper is available? both the &quot;paper&quot; and the &quot;slides&quot; link point to the same document, i.e. the slides.
评论 #14392315 未加载
skybrianabout 8 years ago
Since machines can crash, possibly corrupting data, it seems like you still need replication? The network might be the new bottleneck.
评论 #14391862 未加载
评论 #14451827 未加载
ComodoHackerabout 8 years ago
&gt;Assistant Professor of <i>Databaseology</i><p>Oh, I didn&#x27;t know that&#x27;s a thing.
评论 #14393636 未加载