HackerTrans
TopNewTrendsCommentsPastAskShowJobs

mlerner

no profile record

Submissions

Graham: Synchronizing Clocks by Leveraging Local Clock Properties (2022) [pdf]

usenix.org
59 points·by mlerner·11 miesięcy temu·13 comments

Autothrottle: Resource Management for SLO-Targeted Microservices

usenix.org
22 points·by mlerner·w zeszłym roku·1 comments

Resiliency at Scale: Managing Google's TPUv4 Machine Learning Supercomputer

micahlerner.com
1 points·by mlerner·w zeszłym roku·0 comments

How the anti-cheat is anti-cheating so far

leagueoflegends.com
3 points·by mlerner·2 lata temu·0 comments

ServiceRouter: Hyperscale and Minimal Cost Service Mesh at Meta

micahlerner.com
1 points·by mlerner·2 lata temu·0 comments

A Cloud-Scale Characterization of Remote Procedure Calls

newsletter.micahlerner.com
2 points·by mlerner·2 lata temu·0 comments

A Cloud-Scale Characterization of Google's Remote Procedure Calls

micahlerner.com
1 points·by mlerner·2 lata temu·0 comments

Gemini, Amazon's system for fast failure recovery in distributed model training

micahlerner.com
2 points·by mlerner·2 lata temu·0 comments

Defcon: Preventing overload with graceful feature degradation (2023)

micahlerner.com
237 points·by mlerner·2 lata temu·95 comments

MotherDuck: DuckDB in the Cloud and in the Client [pdf]

cidrdb.org
5 points·by mlerner·2 lata temu·0 comments

Gemini: Fast Failure Recovery in Distributed Training with In-Memory Checkpoints

micahlerner.com
3 points·by mlerner·2 lata temu·0 comments

Gemini: Fast Failure Recovery in Distributed Training with In-Memory Checkpoints

newsletter.micahlerner.com
4 points·by mlerner·2 lata temu·0 comments

XFaaS: Hyperscale and Low Cost Serverless Functions at Meta

newsletter.micahlerner.com
3 points·by mlerner·2 lata temu·0 comments

XFaaS: Hyperscale and Low Cost Serverless Functions at Meta

newsletter.micahlerner.com
3 points·by mlerner·2 lata temu·1 comments

Blueprint: A Toolchain for Highly-Reconfigurable Microservice Applications

micahlerner.com
2 points·by mlerner·2 lata temu·0 comments

Gemini: Fast Failure Recovery in Distributed Training with In-Memory Checkpoints [pdf]

cs.rice.edu
50 points·by mlerner·2 lata temu·13 comments

XFaaS: Hyperscale and Low Cost Serverless Functions at Meta [pdf]

cis.upenn.edu
4 points·by mlerner·2 lata temu·0 comments

Efficient Memory Management for Large Language Model Serving with PagedAttention

newsletter.micahlerner.com
3 points·by mlerner·2 lata temu·0 comments

Efficient Memory Management for Large Language Model Serving with PagedAttention

newsletter.micahlerner.com
1 points·by mlerner·3 lata temu·0 comments

Blueprint: A Toolchain for Highly-Reconfigurable Microservice Applications

micahlerner.com
1 points·by mlerner·3 lata temu·0 comments

comments

mlerner
·2 lata temu·discuss
Author of the paper summary here - my understanding is that XFaaS doesn't run functions that are run in response to user input (e.g. XFaaS does not execute code that fetches and returns data because a user clicked on a button).
mlerner
·3 lata temu·discuss
Good catch! Here's the link to the framework: https://gitlab.mpi-sws.org/cld/blueprint/blueprint-compiler.

Also added it to the blog post :)
mlerner
·3 lata temu·discuss
The x-axis is size of trade, and the y-axis is % of trades at a specific size. ExRates had _many_ large trades, which shows up as a significantly different distribution than other exchanges like Coinbase/Bitstamp/Kraken (which had many small trades, and a long tail of larger size trades). See slide 42-44 here: https://www.sec.gov/comments/sr-nysearca-2019-01/srnysearca2...
mlerner
·3 lata temu·discuss
100% - although this time around there is (in my opinion) _a lot more data_ [1] that regulated markets (like CME bitcoin futures) drive price discovery.

[1] There was data back then too, it just wasn't received.
mlerner
·3 lata temu·discuss
The volume we were looking at was on exchanges (not on chain).
mlerner
·3 lata temu·discuss
It’s Jekyll - and open source! https://github.com/mlerner/mlerner.github.io
mlerner
·3 lata temu·discuss
Neat and funny that they have an uncited figure (the one with the six exchanges with wacky trade size distributions) from research that we did while I was working at Bitwise.

We found 95% of bitcoin trading volume at the time was fake: https://www.wsj.com/articles/most-bitcoin-trading-faked-by-u...
mlerner
·3 lata temu·discuss
https://www.micahlerner.com

I write about new and foundational academic CS research - writing is a way for me to learn and share with others along the way.

Some of my most popular writing is:

- FoundationDB: A Distributed Unbundled Transactional Key Value Store (https://news.ycombinator.com/item?id=28740497)

- Monarch: Google’s Planet-Scale In-Memory Time Series Database (https://news.ycombinator.com/item?id=31379383)

- Ray: A Distributed Framework for Emerging AI Applications (https://news.ycombinator.com/item?id=27730807)
mlerner
·3 lata temu·discuss
Really neat paper - a while ago I wrote a summary of the system: https://www.micahlerner.com/2021/06/12/foundationdb-a-distri...
mlerner
·3 lata temu·discuss
[flagged]
mlerner
·4 lata temu·discuss
Came to the comments and found my blog mentioned. Aw schucks.
mlerner
·4 lata temu·discuss
Author here - thank you for pointing that out! The context in the comment you linked to is also helpful.

In your opinion, would it be correct to remove "uniquely" from the first sentence? Maybe that would help clarify the issue you pointed out. As far as I understand, the CIDs are still unique when referring to _metadata_, and can be used to "unambiguously fetch content from its peers".
mlerner
·4 lata temu·discuss
I wrote a summary of one of the approaches for replacing TCP mentioned in the paper (called Homa) here: https://www.micahlerner.com/2021/08/15/a-linux-kernel-implem...