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.

MOSAIC: Processing a Trillion-Edge Graph on a Single Machine [pdf]

1 pointsby blopeurabout 8 years ago

2 comments

eternalbanabout 8 years ago
Section 2 of OP points us to 5.5 regarding &quot;load balancing&quot;, after reminding us that &quot;Optimal graph partitioning is an NP-complete problem&quot;.<p>In 5.5 we read about <i>single node</i> internal processor load balancing and also the happy comment in the following 5.6 (&quot;Fault Tolerance&quot;) that &quot;in MOSAIC, due to its single-machine design, handling fault tolerance is as simple as checkpointing the intermediate state data (i.e., vertex array).&quot;<p>I would be more interested to hear about processing a &quot;Trillion-Edge&quot; graph distributed over &quot;Two&quot; machines: Partitioning a general graph is not as simple as partitioning k&#x2F;v buckets or relational graphs; and, 10^12 edge order sounds like a lot until you recognize the &#x27;edge amplification&#x27; fact of capturing a logic domain using the weak semantics of named associations.
blopeurabout 8 years ago
Slide deck : <a href="https:&#x2F;&#x2F;taesoo.gtisc.gatech.edu&#x2F;pubs&#x2F;2017&#x2F;maass:mosaic-slides.pdf" rel="nofollow">https:&#x2F;&#x2F;taesoo.gtisc.gatech.edu&#x2F;pubs&#x2F;2017&#x2F;maass:mosaic-slide...</a>