DSys RG Nov 2022: HAT 👒

Paper

Video


Nov 2022 DSys RG: HAT :womans_hat: - Watch Video

Chat Log

00:15:00	Blaise Pabon:	My first meeting and I didn’t do the reading. That won’t happen again.
00:15:57	Brooklyn Zelenka (@expede):	No problem Blaise! Happy to have you along for the ride 🙂
00:16:09	Blaise Pabon:	I’m super grateful to be here. My  Mastodon community (hachyderm.io) is struggling with scaling PostGres so I’m just looking for ideas.
00:17:51	Marc-Antoine Parent:	@Blaise how much did you play with partitioning?
00:18:40	Philipp Krüger:	MAV = Monotonic Atomic View
00:18:59	Zeeshan Lakhani:	A good paper on looking at causal consistent CRDTs and stronger semantics on CRDTs is the cure paper https://dd.thekkedam.org/assets/documents/publications/Report/RR-8858.pdf
00:19:15	Philipp Krüger:	"Under MAV, once some of the effects of a trans-
action Ti are observed by another transaction Tj , thereafter, all ef-
fects of Ti are observed by Tj . That is, if a transaction Tj reads
a version of an object that transaction Ti wrote, then a later read
by Tj cannot return a value whose later version is installed by Ti."
00:19:28	Marc-Antoine Parent:	Thanks Zeeshan!
00:20:13	Philipp Krüger:	Adya
00:20:49	A. Jesse Jiryu Davis:	Adya: https://pmg.csail.mit.edu/papers/icde00.pdf
00:20:57	Zeeshan Lakhani:	On Paxos, Heidi’s dissertation: https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-935.pdf
00:21:23	Zeeshan Lakhani:	Also, PWL talk: https://www.youtube.com/watch?v=Pqc6X3sj6q8
00:24:32	Marc-Antoine Parent:	Of course that’s more likely to be due to A/B testing :->
00:25:28	Marc-Antoine Parent:	No absolute evaluation, but easy to do it relatively, no?
00:31:02	Marc-Antoine Parent:	That’s a great question… is there a good distributed system test fuzzing library?
00:31:36	Blaise Pabon:	+1 @Jesse … it’s all fun and games until someone builds a complex system.
00:31:52	Philipp Krüger:	@maparent I know *of* https://github.com/jepsen-io/jepsen, but have never used it
00:32:06	Marc-Antoine Parent:	Thank you Philipp!
00:32:32	Philipp Krüger:	Although that's perhaps more of a... BFT testing thing?
00:32:33	Philipp Krüger:	not sure
00:32:46	A. Jesse Jiryu Davis:	What's BFT?
00:32:52	Paul Borrill:	Is Kyle Kingsbury's site still a good fuzzing library?
00:32:55	Philipp Krüger:	Oh I meant "byzantine fault tolerance"
00:33:18	aesakamar:	> Kyle Kingsbury's site
00:33:26	aesakamar:	I’ve found the Jepsen stuff really cool
00:33:44	A. Jesse Jiryu Davis:	No Jepsen is not aimed at BFT if I understand correctly, it's a fuzzer that tries to make a system violate its published isolation guarantees
00:34:15	Sodium :	rural situations also seem like a case where availability is both Important and Potentially An Issue
00:34:42	Zeeshan Lakhani:	Yep, Jepson really links to this paper in a way of making sure systems are true to their tradeoffs and choice of tx semantic levels.
00:34:48	Philipp Krüger:	Ok, nice!
00:35:02	A. Jesse Jiryu Davis:	Here's a writeup of a project I did with Jepsen: https://emptysqua.re/blog/python-paxos-jepsen/
00:35:40	Philipp Krüger:	Oh wow. I need to look at this :)
00:35:51	Zeeshan Lakhani:	+1 Jesse. Also, Kyle’s graph on consistency models is one I always go back to, to review: https://jepsen.io/consistency
00:37:54	Sodium :	@Blaise rural as in you physically exist in a not well connected area. like you're on a hike in a remote location or in a cabin with a questionable connection. relying on queuing and retries in those sorts of situation seems not great for UX
00:38:15	Sodium :	but also probably not strictly *avoidable* if you don't have the data locally, so
00:38:28	Blaise Pabon:	Ok, so , like my old house in the Santa Cruz Mountains.
00:40:16	Blaise Pabon:	Ooooh, I had not considered how to represent distributed infra in vitrina.
00:40:20	Blaise Pabon:	https://vitrina.readthedocs.io/en/main/howto/consume.html#how-does-vitrina-work
00:40:22	Zeeshan Lakhani:	“was thinking about Paxos as I rode the subway home late last weekend, and as I walked the final blocks in the cold from Union Square to my apartment, suddenly it all fit together. ‘Yes, Leslie,” I thought, “you’re right, it really is simple.’” Haha Jesse. You’re making me reminisce about NYC now; also, great line.
00:40:41	Blaise Pabon:	Thank you @aesakmar
00:43:11	Emma May:	Oooooo, I'm gonna have to check out vitrina more! That seems super relevant to some stuff I'm working on right now
00:44:33	Zeeshan Lakhani:	So, Cmeik and Heather Miller ran this workshop: https://2017.ecoop.org/series/pmldc, https://2016.ecoop.org/track/PMLDC-2016 for a couple years
00:45:08	Zeeshan Lakhani:	With many of the ideas linked to this https://arxiv.org/pdf/1701.07615.pdf.
00:46:18	A. Jesse Jiryu Davis:	New Directions in Cloud Programming: https://arxiv.org/abs/2101.01159
00:46:42	Zeeshan Lakhani:	And video for cliff notes: https://www.youtube.com/watch?v=lY-_IKGvITU
00:50:41	Sodium :	https://www.unison-lang.org/
00:50:56	Paul Borrill:	Physicists would disagree that (a) centrally referencable clocks, and  causality are not achievable in Spacetime.  See https://itsabouttime.club
00:54:00	Marc-Antoine Parent:	https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/distrmm-long-3-25-13.pdf
00:55:24	Blaise Pabon:	Donald Knuth is still working on P vs NP
00:55:57	Zeeshan Lakhani:	I’ll link this elsewhere, but a notion for listing papers for 2023: https://www.notion.so/fissioncodes/Distributed-Systems-Reading-Group-2023-epoch-265f597d737243e08862e7cac2601d7e (hope this works!)
00:57:58	Paul Borrill:	Anyone  want to collaborate on writing a deep-thinking paper?
00:58:58	Philipp Krüger:	fission.codes/discord/ I think?
00:59:03	Philipp Krüger:	ugh https://fission.codes/discord/
00:59:12	Zeeshan Lakhani:	For unison, the chez link pr https://github.com/unisonweb/unison/pull/3589
00:59:32	Sodium :	that link works for me
00:59:45	Brooklyn Zelenka (@expede):	#dsys-reqding-group channel
00:59:48	Brooklyn Zelenka (@expede):	* reading
01:00:08	Sodium :	fork!
01:00:10	Philipp Krüger:	fork what you can't!
01:00:11	Sodium :	fork what you can't
01:00:13	Philipp Krüger:	:D

Via @emptysquare in the Discord