mrmaddog

Arista Networks | San Francisco, Vancouver | Full-time, onsite

We're building a modern network monitoring system and are looking for motivated, creative engineers. Let's chat if you - are an experienced Javascript developer (bonus points for experience with the React ecosystem) with an eye for great data visualizations - or, love streaming data processing and you'd be excited to work with HBase, Kafka, and Golang.

http://arista.com | ryan@arista.com

If you don't have a Wall Street Journal subscription, a Google redirect link might help [1]. Alternatively, here are a few choice quotes from this biting WSJ editorial:

  Is anyone in charge at the Justice Department, or are
  junior prosecutors running the joint?

[...]

  Such assertions were as false in Brooklyn as in San 
  Bernardino. Two hours and a half before a deadline on Friday 
  night, the government withdrew the case after “an individual 
  provided the passcode to the iPhone,” according to legal 
  filings. This second immaculate conception in as many months 
  further undermines the FBI’s credibility about its 
  technological capabilities. Judges ought to exercise far 
  more scrutiny in future decryption cases even as Mr. Comey 
  continues to pose as helpless.

[...]

  Yet forgive us if this “conversation” now seems more like a 
  Jim Comey monologue. The debate might start to be productive 
  if the FBI Director would stop trying to use the courts as 
  an ad hoc policy tool and promised not to bring any more 
  cases like the one in Brooklyn.

[...]

  Meanwhile, the White House has taken the profile-in-courage 
  stand of refusing to endorse or oppose any encryption bill 
  that Congress may propose. If the Obama team won’t start 
  adjusting to the technological realities of strong and legal 
  encryption, they could at least exercise some adult 
  supervision at Main Justice.

[1]: https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&c...

I'd disagree with point 4. Their Instagram feed has all of the stunning beauty that National Geographic originally became famous for: https://instagram.com/natgeo/, and commands a 35.4 million person following. I'm not sure how you parlay this into a "selling relationship", but it is a great way to outreach to a younger generation.

Nope—just flip the upper left coin (000000), as that has no effect on any of the parities. Flip the bottom right hand corner if you need to switch all the bits.

I know Show HN's are usually reserved for small, personal projects, but I'm pretty excited to finally make EOS SDK public. Essentially this gives you Python and C++ bindings to write apps that hook directly into Arista's networking stack. Also, since EOS is Linux (Fedora), you can use your standard environment and debugging routine during development, then test on an EOS VM (called vEOS), all of which keeps the barrier to getting started a lot lower.

We're still expanding the number of modules that are available, but already we've seen agents ranging from simple event-driven monitoring, to full-blown controller-driven custom routing protocols. It's pretty awesome to see what people can do when you give them a proper interface to the networking stack!

I really like the idea of unobtrusive anomaly detection as a product, but I wonder how much benefit I could actually derive from this.

Perhaps it is nice to have a log of whenever the front door opens, but that feature certainly won't sell me on its own.

It is cool that they can tell when a window breaks, but... how much good does that notification do for me? How confident can I be that this is not a false positive and immediately call the police? Or do I rely on the "soft alarm" to hopefully scare away the would-be intruders?

I don't see what the "light up" option affords. If I'm there, I'll know that loud music is playing. And I don't get the sense that this is meant to replace my smoke alarm/carbon monoxide alarm, so at best the light is an early warning system that I should open a window to vent out some smoke from my cooking before the real alarm sets off.

Overall, this seems like a really cool concept that has a lot of marginal benefits and zero killer features. Hopefully that sell-story will be embellished upon in the real kickstarter!

I've used Hitch regularly over the past few months, and it works well for short trips. Half the time I don't carpool (and I still pay half price), and the other half of the time adds just a couple extra minutes (haven't gone more than 5 minutes out of my way). My one issue with Hitch so far is that occasionally it takes a while to get me a ride. Uber has a leg up here, as it has a huge scale advantage. When I open my app, I can see 5 ubers in a 3 block radius - which is pretty astounding reach.

Arista Networks - Software and Hardware Engineers

[Santa Clara, with satellite offices in San Francisco, Vancouver, Bangalore]

Tags: Networking, Distributed Systems, Operating Systems, SDN

We're shaking up the world of datacenter networking (think switches and routers), led by an incredibly smart and pragmatic engineering team. If you're on the software side, work on a modern software stack (amusingly, that's a rarity in the networking world), and hack on incredibly distributed systems approaching ridiculous scaling requirements, mostly in C++/Python. The hardware side builds some of the meanest and most efficient boxes out there—the things that they churn out of their workshop never ceases to amaze me.

In any case, we're constantly shipping code, genuinely working to improve the state-of-the-art, changing the entire networking industry, and making money while doing it (See: $ANET). It's a pretty fun time - come check us out at http://www.arista.com/en/careers/engineering

Feel free to ask me any questions, too (contact info in my profile).

I fail to understand your insinuation here - in the future an articulated reply would be preferable. In each of the comments, I read rayiner's argument as "Your metadata isn't your data, and thus cannot be considered your papers and effects." My personal emails stored on GMail's servers seem to be squarely in domain of "papers and effects," according to the pages rayiner refers to, but does not offer any insight on whether the metadata (who I emailed) should be in this domain.

I believe rayiner refers to exactly this issue in the above comment when calling out the Third Party Doctrine.

Sorry for the deleted comment - after circulating this article internally, I learned Arista already has better solutions for this particular problem than writing handlers to interface statuses. Decided to retract my comment (while you were replying, incidentally) as there are a number of solutions you can implement on a switch involving BGP knobs, fast-server failover, ECMP and consistent hashing.

I've been pleasantly surprised by some of Buzzfeed's journalism. This article [1], a rich narrative following one of the few female boxing judges, has original reporting and is well written. Funnily enough, I discovered this article via a link from New York Times' wonderful new NYT Now app.

[1] http://www.buzzfeed.com/joelanderson/the-woman-who-kod-manny...

Seems like I'm not the only one that can't get access on-demand. For the curious, here is what the site looks like: http://cl.ly/image/0a3V1d2W1c3J (Each group of three lines in the top bar graph represents one day.) It also lets you know if your tweet attains an unusually high "reach" is: http://cl.ly/image/1p2X3b0x2X1S - which corresponds to the number of feeds your tweet showed up on via retweets.

My personal account has had this for a few months, but my friendly twitter bot (@8ball_) still does not have access. Is there some sort of gatekeeping mechanism? The help page refers to twitter card analytics, but that is not helpful. It would be nice to roll my own statistics off of that CSV instead of relying on hodge podges of services like Wildfire (https://monitor.wildfireapp.com/count_reports/display?twitte...)

In any case, I'd love to see this roll out to more accounts: I think the information on engagement, follow:unfollow ratios and mentions over time are incredibly useful for creating more meaningful content (for businesses and automated services), though I am not convinced such behavior is good for individuals, and am curious whether access to a dashboard increases or decreases engagement.

I thought this was going to be vaporware, but man, I'm impressed. If it can really do half of what the introduction video shows, I'm can't wait to play with it!

One thing I think is particularly interesting is that the onboarding process lets you use natural language to figure out what WolframCode to write. Slick!

Somewhere between Jupiter and Saturn, I had to resort to

   $(".essay").each(function(i){console.log($(this).text())});

to read the fun little text snippets. I like how it got fairly philosophical—everything eventually turns to philosophy if you are stranded out in space. Thanks for putting this together; it was a nice take on the perspective of our universe.

Great story! One thing that stuck out to me—if you had an infinitely powerful computer, the "coding" aspect of it should take no time at all. All you have to do is come up with a description language to say what you want•, and b) seed rules for a genetic programming language, and then throw it at your infinitely powerful computer. You'll get a working program instantly.

• This might still be significantly difficult. If only you had a simulated version of yourself that you could use as a control subject so the computer could test whether or not the generated program met your expectations...

Some interesting questions and answers:

   [Q] Dear Mr. Metcalfe, First, thank you for all your brilliant work over the
   years. My question is how do you feel about the NSA collecting everything?"

   [Metcalfe] America needs an NSA, because we have dangerous enemies. Perhaps
   they've been over-doing it lately. I am worried that the current administration
   will destroy NSA in the current scandal.

It is always interesting to hear a different point of view from a person in tech. Also love his sense of humor:

   [Metcalfe] Ethernet cables should be yellow. And that's that.

I have not yet seen "Her", but this strongly reminded me of Ender's communication with Jane from the "Ender's Game" sequels. One of the most interesting facets to their conversations is that Ender could make sub-vocal noises in order to convey his points—short clicks of his teeth and movements of his tongue—that Jane could pick up on but humans around him could not. It is the "keyboard shortcuts" of oral communication.

If "Her" is really the future to HCI, then sub-vocal communication is a definite installment as well.

Once you do decide to use an assert in python, there are a couple things to consider. The first is to make sure you are expressive in your message. Your assert may very well hit one day, and it may only occur at the most unpredictable times, so you'll want to make sure you have enough information to see why the assert failed. To use the OP's example:

  assert x > 0, "x is not zero or negative"

When this fails, you're going to be at a loss to understand what actually happened. If you change the message to something like the following, you're going to have a much easier time tracing through your program to understand why your constraint was violated:

  assert x > 0, "Expected positive x, got: %d" % x

Once you get in the habbit of this, you'll quickly run up against style-guide imposed line limits. My usual trick here is to use parenthesis and let python's automatic string concatenation work its wonders, but you have to be careful because

   assert (x > 0, "This is my longer message "
           "for x")

evaluates to a tuple, and thus is always true. Instead only use parens around the message:

   assert x > 0, ("This is my longer message "
                  "for x")

Lastly, don't use asserts in tests. Use the standard unittest library which will do a much better job explaining what was received, what was expected, and what the difference between them is.

Well this is silly. Stanford graduation rates have been increasing over the past 10 years: (http://ucomm.stanford.edu/cds/2012 - search for point B11). 1998-2003 were all in the 90-93% range. This last year had a graduation rate of 95%. This doesn't support the opinion that students are dropping out 'en masse.' As for "professors having a financial stake in their student's work" – since when was this not the case? Professors have been dependent on grant money based on the work of their student researchers for decades; only previously, the students had no way of making money off of the work themselves.

I do think that it would be good to establish and maintain a set of ethics for professors who have a financial stake in companies, but it seems like this piece was more of a ranting projection built for page views than a thoughtful criticism on the state of the university.

Edit: And I should have researched a little more: Stanford does have a policy regarding faculty investments: http://doresearch.stanford.edu/policies/research-policy-hand.... Points 3 & 4 seem particularly applicably as a response:

   3. Faculty must foster an atmosphere of academic freedom by 
   promoting the open and timely exchange of results of 
   scholarly activities, and ensuring that their advising of 
   students (defined for this policy to include postdoctoral 
   scholars and other trainees) and their supervision of staff 
   are independent of personal financial interests. Faculty 
   should inform students and colleagues about outside 
   obligations that might influence the free exchange of 
   scholarly information between them and the faculty member.
   
   4. Faculty may not use University resources or personnel, 
   including facilities, staff, students or other trainees, 
   equipment, or confidential information, except in a purely 
   incidental way, as part of their outside consulting or 
   business activities or for any other purposes that are 
   unrelated to the education, research, scholarship, and public 
   service missions of the University.

I like how you can click on a person's representative dot and follow them through the different sorting metrics. One point stood out to me: how does Assistant Treasurer Ms. Collier get $289,534 in "other" pay? The raw data provided didn't offer any details, and http://www.mercurynews.com/salaries/bay-area/2012 classifies "Other" as lump sum payouts for vacation, sick-leave, bonuses and comp-time. How does the assistant treasurer get such a high bonus? (For reference, one other assistant treasurer was included in the data, and made a base of 150k (vs Ms Collier's $30K), but only 6k in "other" payment).

Visualizations make it very easy to spot outliers like this.

Ah, I didn't realize this was done as a manual process back then. I assumed the engineers were working on an automated system to do the segment tuning for them, and that system was the new "feature" that Mr. Gates thought the engineers were wasting too much time on. Feel free to re-read my above comment with "segment tuning" replaced by "some other feature." Thanks for the correction.

Not exactly. I am a huge fan of code modularization, but what modularization offers you is the ability to decrease the exponent (i.e. to be devTime^1.01 instead of devTime^2). Some features though (including segment tuning, from the sound of it), will touch a ton of other features regardless of how well you modularize your code.

I work at a networking company (Arista), and a lot of the interesting problems come from this sort of interaction. Our entire OS was built so an agent would be resilient to changes in another agent, and this modularization means there is very little "spaghetti code". However, when you are building a feature (say, a new routing protocol), you have to be extremely conscious of how it interacts with everything else: various reliability features (i.e. Stateful switchovers), configuration features (saved configs, public APIs), resource contention (TCAM utilization), etc. etc. If that new routing protocol was the first thing we implemented on our switch, it would be a complete breeze. In the context of other features though, this becomes a more intensive project (though in codebases without proper modularization you'd find this task "herculean" as opposed to "intensive").

I think the real lesson here is that every time you add a feature, you are adding development overhead to every other feature, leading to exponential development time. I'm sure if segment tuning was one of the first thing to go into windows, it would have been much simpler to implement. However, at later stages in the development you have to account for how different, already-written applications are loading their memory, and thus the feature needs handle far more edge cases than it would have if this was an earlier-implemented feature. I'm not saying the overall time to develop segment-tuning would be shorter, in the long run had it been developed first, but the perception of development-time for this feature would be shorter (since any segment tuning edge cases that you ran into while writing an application would be considered part of the app's development time, not segment-tuning's development time).

I'm sure that if Mr. Gates had implemented FAT at a later date, he would have needed a much longer plane ride.

Here is a link to the live demo: http://nlp.stanford.edu:8080/sentiment/rntnDemo.html

This is really fun to play with, and I'm surprised how well it can parse the sentiment of sample sentences I threw at it. I've tried a couple random examples (like "I don't know what the artist was smoking, but the song made no sense (though I liked the beat!)") and have not yet gotten a wrong analysis. Even the phrase parsing is pretty spot-on.

As a side note, this is much more interesting than the "sediment" analysis I excepted after skimming the title. (Unfortunately though, the analyzer got this final sentence wrong: http://cl.ly/image/301u1q46263m)

Edit: seems like this system could get significantly more robust with more data. If you look in the comments section, you can see some comments from the professor himself, i.e. "Possibly because the word "buying", only appears once in the entire dataset and it's in a pretty negative context: http://nlp.stanford.edu/sentiment/treebank.html?w=buying"

If you gave it 100,000 phrases, I wouldn't be surprised if it could hit the 95% mark that Socher mentions.

I really enjoy Gruber's holistic reviews of Apple's new devices. He doesn't attempt to be Anandtech (whose reviews I found thoroughly engrossing and well written), but rather he paints a picture of how the device feels and how it fits into Apple's larger strategy, which I find equally important.

One interesting thing I've discovered about Gruber's columns is that you often have to read between the lines and get accustomed to how he presents his narrative. Praise for Apple is freely handed out, weaknesses are noted, and praise for other companies is guarded (when said company is in competition with Apple). For example, he notes the weaknesses of Siri in a generally positive sentence, and states that "Google Now is faster." I think he would agree that Google Now is not only faster, but also provides better results with better comprehension, and really the only downside is that you can't access it with a long-hold on the home button. That's not something I'd expect Gruber to say. However, once I recognized how he couches his concerns, I've found that he has some of the most insightful, nuanced and thoroughly processed views about Apple overall.

Why is harvesting fingerprints such a catastrophic possibility? The government already has your fingerprints if you've ever left or entered the country, and they've likely been taken during other periods in your life as well. Hell, even the Cambodian government decided to scan my fingers when I passed through. DrCube's counter-argument [1] is logical: that with these existing databases, the government can now access your phone. But really, is this feature being touted as a impenetrably secure way to unlock your phone? It seems to be positioned as "a way to make accessing your phone convenient for you and non-trivial for others," and I think this feature does exactly that.

If you can't trust the phone vendor, why are you not more paranoid about entering your username/password on the device? What about location data? I really can't fathom why decrying fingerprint technology is the nexus of your argument here.

https://news.ycombinator.com/item?id=6364725

Sounds like something VXLAN is well equipped to handle - VLANs tunneled over IP to create private networks in large cloud environments: http://en.wikipedia.org/wiki/Virtual_Extensible_LAN I wonder if DO will be using something like this to mitigate direct attacks (though VXLAN wouldn't solve the problems of link oversubscription—but there are other mechanisms in place to solve this)

What are your thoughts on joining a big company vs joining a startup (as opposed to starting a company yourself)? On average, it seems one would still learn faster and be able to take on many new roles as an employee at a startup, but an employe also is in a chaotic world that he or she doesn't completely control (and often working with incomplete information).

Do you think this trade-off is worth it? And, since I'm guessing you lean towards 'yes' on that question, do you think there are different questions a prospective startup-employee should ask as opposed to the questions a potential investor in the startup would ask?

You can also drag any file or folder onto your terminal, which will paste the absolute filepath at your cursor point.