Neil Ramaswamy

249 posts

Neil Ramaswamy

Neil Ramaswamy

@entersudonym

Applied research @OpenAI; previously streaming @Databricks, shenanigans @BrownUniversity

Katılım Ocak 2020
228 Takip Edilen612 Takipçiler
Sabitlenmiş Tweet
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
The world is becoming real-time. A batch job on a cron *will* burn you. Watch the video to deep-dive into how real-time SQL engines function correctly and with low-latency!
Phil Eaton@eatonphil

Here is @entersudonym speaking about: Correct and low-latency real-time streaming joins. Recording from last Thursday's NYC Systems Talks. youtube.com/watch?v=I6MJqN…

English
1
2
48
13.3K
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
The SF Systems Club is closing out 2025 with a fun one: @AntithesisHQ will speak about building a deterministic hypervisor. They mock clocks, run a custom Linux fork that plays nicely with hardware caches, and pull a bunch of other sneaky tricks :) luma.com/sdmlmjcp
English
1
1
11
1.5K
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
in openai confusing naming tradition, I pronounce our new open source model as “gee-pee-tee-for-o”, as in “gpt for *o*pen source,” much to my colleagues’ dismay. it’s a great model, enjoy!
English
2
1
18
2.5K
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
@ShriramKMurthi one of my core memories from PL is you saying something along the lines of "be who you are, i won't judge you, unless you implement a language with dynamic scope, in which case i'll DEFINITELY judge you."
English
1
0
8
285
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
The SF Systems Club is back in May, hosted generously by @OpenAI! Brian Lagoda from @AntithesisHQ will talk about building intelligent fuzzers, and Will Noble from Google will talk about WASM components! (Link in replies)
English
2
7
134
14.8K
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
Wow! The SF Systems Club—hosted yet again by @databricks—is on a roll! We got almost 200 signups in just a few hours. (Finally, the SF Systems Club is branching out from data systems into formal methods and security 🙂)
Shadaj Laddad@ShadajL

The SF Systems Meetup is back! On 2/27, we're excited to have headline talks from the creator of FizzBee and a research collaborator with Signal. This is going to be a super fun night diving deep into making distributed protocols work, hope you'll join us! lu.ma/vqjf30k3

English
0
2
15
4.2K
Colt McNealy
Colt McNealy@coltmcnealy·
There have been a half-dozen surprises about small stylistic things that are broken on our docs after we added Tailwind CSS. I could never be a frontend engineer; there is no rhyme or reason to any of it.
English
1
0
5
258
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
@davidtgoldblatt @geofflangdale Hm, why is little endian "objectively right" here? Intuitively, I'd think that when thinking about raw bits, the natural order is big endian.
English
1
0
0
43
David Goldblatt
David Goldblatt@davidtgoldblatt·
@geofflangdale Bitmap interfaces maybe? Having to force producer/consumer to agree on a word size is sort of annoying in various contexts, and little-endian is sort of objectively "right" there.
English
1
0
3
160
Geoff Langdale
Geoff Langdale@geofflangdale·
So this is likely a tired debate, but... are there *any* objective advantages of big-endian vs little-endian? I vaguely like the fact that if you have a variable-sized structure in LE you can load it in from LSB to MSB and mask off the bits that aren't needed, but this feels ...
English
6
0
19
2.4K
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
@Jaidchen @badcop_ Oops... the example was supposed to be with 0 😅. Since clearly I can't do this from memory, here's the original link where I saw this: #comparing-null-to-0" target="_blank" rel="nofollow noopener">github.com/denysdovhan/wt…
Neil Ramaswamy tweet media
English
0
0
0
67
badcop
badcop@badcop_·
do the "javascript makes no sense" people have any examples other than 2 + '2'
English
134
21
1.4K
216.4K
Colt McNealy
Colt McNealy@coltmcnealy·
What is "Left" in scala?
Colt McNealy tweet media
English
1
0
0
237
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
// This function does asdfwihhueksadfj def asdfwihhueksadfj() = { ... } (Thanks for the comment! I couldn't tell!) APAITPWMYCKWYL&IAVP h/t @ShriramKMurthi
English
1
1
4
1.6K
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
@vanlightly Yes, exactly. Currently, most customers just use foreachBatch and write the merge themselves, with their knowledge of what the primary key is. This works well with micro-batching, but for Real Time mode (record-by-record processing), that approach doesn't work.
English
0
0
2
58
Jack Vanlightly
Jack Vanlightly@vanlightly·
@entersudonym It seems like something Spark Structured Streaming should be able to do, but I don't think it supports it right now.
English
1
0
0
81
Jack Vanlightly
Jack Vanlightly@vanlightly·
So it seems that no compute engines have a streaming sink capable of updates and deletes with Delta Lake, just appends?
English
4
0
12
2.2K
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
@vanlightly The streaming sink, though, could track the primary key through the plan and use that in the merge automatically, no? For a single aggregation/join, this is doable, but for arbitrary code (ProcessFunction, FlatMapGroupsWithState), you'd need the user to specify it explicitly.
English
1
0
1
72
Jack Vanlightly
Jack Vanlightly@vanlightly·
I think it comes down to Delta (and Iceberg) not baking in primary keys into the table format itself. So can't do upserts natively (like Hudi and Paimon). Gotta resort to the append-only table + merge pattern.
English
1
0
5
655
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
@ngeloxyz One time my work Slack status was "I'm not just customer obsessed—I'm obsessed with the customer." Didn't get fired. Maybe they liked my attitude. Or maybe HR just didn't see it. Please nobody tag Databricks in the replies. I don't want to find out
English
2
0
2
198
Angelo 🇵🇷
Angelo 🇵🇷@ngeloxyz·
when people say they are customer obsessed they aren’t about it like me I keep dossiers on my prospects, I put their LinkedIn profile photos on my office wall, my living room, I know who my target buyer is, I met them, at the grocery store- they really don’t like Bill com so i-
Angelo 🇵🇷 tweet media
English
2
0
10
704
Neil Ramaswamy
Neil Ramaswamy@entersudonym·
At the brewery after NYC Systems yesterday, I mentioned the typo in the linearizability definition—found ~30 years later—to @eatonphil and @nikhilbenesch! The first author's talk from PODC 2021 explains the original and fixed definition remarkably well: youtube.com/watch?v=aOnIHq…
YouTube video
YouTube
English
0
1
13
2.2K
Pooja
Pooja@algurlrhythm·
@ShadajL @databricks Wonderful! Would the talks be recorded for on-demand watching?
English
1
0
1
159
Shadaj Laddad
Shadaj Laddad@ShadajL·
The SF Distributed Systems Meetup (now under the SF Systems Club umbrella) is back! Join us 8/29 at @Databricks SF for an exciting evening featuring talks by Ben Hindman (co-creator of Mesos) and community members! lu.ma/x17bg5fh
English
4
13
68
23.6K

Keşfet

@AntithesisHQ @ShriramKMurthi @nilskch @eatonphil @OpenAI @AmplifyPartners @databricks @coltmcnealy