grogers

30 ms·

301.
by grogers 11y ago
Thanks for linking that. What surprises me is that in their simulation which they base the results on - the early 10 years are way different than the later 10 years. The early years show fairly steady growth (although most of it is concentr
302.
by grogers 11y ago
Heartbeating is only used for failure detection. Because raft is a consensus mechanism, failure detection is allowed to return false positives. In practice false positives cause a blip in throughout as a new leader gets elected, but since t
303.
by grogers 11y ago
Just from skimming the design doc, it appears that if your clock skew exceeds the maximum bound, it would break linearizability. I haven't parsed through all the details of their SSI implementation, but it appears that even with arbitr
304.
by grogers 11y ago
My analog to this is when someone claims that partitions are rare so you can ignore them, any product they make is very likely to lose data during partitions.
305.
by grogers 12y ago
The difference is the scheduling. In erlang, processes are preemptively scheduled, in akka they aren't. Erlang goes to great lengths to avoid one process from starving another. With akka you have to go to some lengths yourself to preve
306.
by grogers 12y ago
This article also makes it even more complicated by calling the C in CAP "atomic consistency". While technically correct as that means the same thing as linearizability, that term is rarely used in comparison. When comparing the o
307.
by grogers 12y ago
The TLA+ tools can do exhaustive model checking, as well as mechanically checked proofs. The proofs typically use an SMT solver to verify them. Of the two I've only used model checking in a serious way, but you can find Lamport's
308.
by grogers 12y ago
Looking through the pages for "software engineer" I see mostly in the range of $90k - $135k with a few outliers. This is pretty reasonable for a big software company in my experience. There are some that are $200k or more, but the
309.
by grogers 12y ago
Do the branches happen at run time or are the optimized routines selected at install time (or something else?)
310.
by grogers 12y ago
It probably depends on where you work. Most of my colleagues are amazing developers, but very very few of them actually do side projects at home. Most of us wish we had more time, and would like to do this or that as a side project - but th
311.
by grogers 12y ago
This seems roughly the same as the monopoly question. Both are seemingly more about enumerating all the edge cases than anything to do with software design or implementation. In 20-30 minutes what do you generally expect to get out of askin
312.
by grogers 12y ago
AFAIK only spanner uses real time as a core primitive. They don't use ntp ;-) Monotonic sequence numbers loosely based on real time are common, since you can correlate the sequence number back to an actual time to look at logs, etc. Ti
313.
by grogers 12y ago
You can't make a durable write to disk in 15us. SSDs are around 2 orders of magnitude slower than that. 15us is about right for in memory on two nodes in a single data center though. Most people wouldn't consider that durable thou
314.
by grogers 12y ago
It does use ChaCha20, and the implementation will use something else in another 20 years most likely. http://www.openbsd.org/papers/hackfest2014-arc4random/mgp000...
315.
by grogers 12y ago
That probably came across wrong. What I meant was that the ability to do concurrent access is the important thing, not the performance of it. CPUs aren't any good at contended access to memory cells, but they still allow it and it work
316.
by grogers 12y ago
I'm probably the minority on this, but I think innodb's docs on its isolation modes are clearer and easier to understand than postgres'. That could be because I have internalized them over many years though. To me, the only s
317.
by grogers 12y ago
They aren't mutually exclusive. You can't modify the same row concurrently and have serializability, but that isn't a common requirement. The common requirement is that you can have concurrent transactions operating on differ
318.
by grogers 12y ago
I thought the main endurance metric for spinning disks is total bytes read or written? AKA time where the head is right up close to the platter. That doesn't seem too far off to track both total read and write IOs and/or bytes.
319.
by grogers 12y ago
Except for "static binaries" (which isn't clear what benefits you desire out of them), java has all those things. The other choices are certainly there.
320.
by grogers 12y ago
First off, you can't have CA in a distributed system, but I'm guessing that was a typo and you meant CP or AP. Given that it is based on spanner, and claims to have something close to linearizability, it is CP.
321.
by grogers 12y ago
Keep in mind that they define DC := AWS region Rack := AWS AZ That should tell you that they need multi-region availability, something that DynamoDB will likely never provide.
322.
by grogers 12y ago
You still need ordering amongst writes. Without that you will get into a situation where there is no majority, all three nodes have different data.
323.
by grogers 12y ago
I doubt any consensus algorithm can be simpler than single decree paxos. But what people usually care about is deciding on a sequence of operations. Paxos made live, and others have engineered their own solutions to make multi-paxos efficie
324.
by grogers 12y ago
We used an rdbms for our workflow engine (what you are calling jobs) and it works great. We needed to react to state changes and process workflows within seconds, with dozens of workers executing different jobs in parallel. This is well wit
325.
by grogers 12y ago
The whole industry has been slowly trending away from eventual consistency. Google proved with spanner that it is possible to do linearizable transactions at mega scale. Even Amazon's DynamoDB offers strong consistency for individual r
326.
by grogers 12y ago
Big torque from an engine sounds impressive but in reality only power matters. You can always change gear ratios to get more torque, which means more force to the ground. Power determines how fast you will accelerate, how fast your top spee
327.
by grogers 12y ago
Can someone explain to me how Raft differs from Viewstamped Replication? From reading both papers (vr revisited) it looks like Raft just renamed all of VR's nomenclature without changing anything significant. Paxos is fairly different
328.
by grogers 12y ago
Racing proposers cause paxos to fail to converge. A: Prepare(1) B: Prepare(2) A: Accept(1, Va) # fails. A: Prepare(3) B: Accept(2, Vb) # fails. Etc... In practice randomized backoffs when this happens causes it to converge extremely fast, b
329.
by grogers 12y ago
IIRC there was a similar rumor that S3 reduced redundancy storage was just cheaper rates for the exact same service. Not sure if that is true, but certainly meets the minimum viable product. Glacier as S3 with sleeps seems to be a pretty re
330.
by grogers 13y ago
This seems orthogonal to me. Tcold for the efficiency is the exhaust temp, not the intake. For pure efficiency, it is actually beneficial to capture waste heat in the exhaust to heat the intake air. But it destroys your power density, for t

More ›