Brandon Holt

Apple

Mon, 10 Oct 2016 00:00:00 -0700

I'm now a System Software Engineer at Apple, building tools to help engineers.

Graduation

Wed, 17 Aug 2016 00:00:00 -0700

Successfully defended last Friday and submitted the final version of my dissertation! Thank you to everyone who has helped me along the way! I'll be starting at Apple down in Cupertino in October!

Socc Sosp

Mon, 01 Aug 2016 00:00:00 -0700

Our paper on Disciplined Inconsistency will appear at SoCC 2016 in October, and other work I've been involved in on the Diamond mobile/cloud data synchronization system will appear at OSDI 2016!

Cse351 Sp16 Done

Fri, 10 Jun 2016 00:00:00 -0700

Just finished my first quarter of teaching, CSE 351: The Hardware Software Interface. Had a fantastic time, thanks to all the help from TAs, previous instructors, and my students!

Ipa Tr

Mon, 06 Jun 2016 00:00:00 -0700

Our full-length paper on Disciplined Inconsistency, currently in submission, is now available as a tech report.

Generals

Tue, 15 Dec 2015 00:00:00 -0800

I passed my Generals exam and am now a real PhD candidate! The unusually curious can check out my generals document on Mitigating Contention in Distributed Systems.

Teaching Next Winter

Tue, 04 Aug 2015 00:00:00 -0700

I am officially teaching CSE 351 (Hardware/Software Interface) next ~~Winter~~ Spring quarter. This is the first time you get to see end-to-end how computers actually run programs — so excited to teach it!

LLVM Debugging Tips and Tricks

Mon, 03 Aug 2015 00:00:00 -0700

Back when I was working heavily with LLVM, I learned a bunch of little tricks that made my life so much easier. I meant to document them back when they were fresh in my mind, but didn't get around to it. Now recently I've been chatting with several colleagues at UW that are just getting started with LLVM, and thought I'd go back and put together a couple of the useful tricks I learned.

Disciplined Inconsistency

Sun, 03 May 2015 00:00:00 -0700

In order scale services with tight latency (and throughput) requirements, distributed systems programmers are forced to accept stale or inconsistent data. Typically this means using an eventually consistent data store like DynamoDB, MongoDB, or Riak. With this weaker consistency model, it's hard to know when writes will become visible to others, or what state of the world you're going to get when you read.

Musings from PaPoC

Wed, 29 Apr 2015 00:00:00 -0700

Papoc Musings

Eurosys

Fri, 24 Apr 2015 00:00:00 -0700

Presented Claret at PaPoC workshop and EuroSys poster session. Got some fantastic feedback, watch for followup posts as I sort through it all.

Oopsla Artifact

Fri, 13 Mar 2015 00:00:00 -0700

I will be on the 2015 OOPSLA Artifact Evaluation Committee.

Scholarly markdown (with R) for fun and (non-monetary) profit

Tue, 24 Feb 2015 00:00:00 -0800

I've been thinking a lot recently about how I can making it easier for me to share ongoing research.

Simulating an interactive Twitter workload

Wed, 18 Feb 2015 00:00:00 -0800

This week I submitted some early work on mitigating contention on hot keys in key/value stores to a workshop co-located with EuroSys. In order to better motivate the work, I wanted to recreate scenarios from real life that lead to lots of contention on a few keys. In particular, what is most interesting for my work is cases where a lot of clients are all attempting to read and update the same key at once. One place where this happens organically is when something, such as a tweet, "goes viral" and is suddenly being seen, commented on, and re-shared by millions of people. This web phenomenon causes all kinds of problems for designers and maintainers of social networks or really anything on the internet. Though Twitter's failwhale has been retired for a couple years now, and Twitter can now withstand high tweet volumes, even the spikes that follow a #WorldCup goal or Katy Perry's SuperBowl shark, some situations can still slow it to a crawl, like Ellen DeGeneres's 2014 Oscar selfie.

A plan for un-crashing CrashPlan

Thu, 13 Nov 2014 00:00:00 -0800

CrashPlan is by far my favorite tool for backing up my computers. I'm never able or willing to plug my laptop in to my backup harddrive for long enough for Time Machine to do its job, all of attempts to make Time Machine work over the network (proxied through my always-on desktop) failed. I'm not rich enough to go around buying Apple's Time Capsules or Airport Extremes (I save my money to buy all of Apple's other hardware). CrashPlan, on the other hand, has been a great solution because it actually does a good job of backing up over the internet and I can backup between machines that I own at different sites for free. However, I've been having problems getting the CrashPlan UI App working for a while.

⚗ Alembic

Sat, 16 Aug 2014 00:00:00 -0700

One of the hardest parts about writing distributed applications is locality: where to put the data, how to access it, and where to perform computation to minimize how much data must be moved. My new paper, to appear at OOPSLA'14 in October, describes a compiler analysis called Alembic which helps distributed application programmers deal with locality by making threads migrate to the data they use automatically.

Git Reflog

Wed, 12 Feb 2014 00:00:00 -0800

This is actually a really simple tip. But I went quite a while using Git before learning this secret. The moral of the story is that Git, while it has one of the worst, most unintuitive command-line interfaces ever, really does save almost everything (okay, the one exception I can still think of is doing a git checkout and losing uncommitted changes).

Quick LLVM Tricks

Thu, 30 Jan 2014 00:00:00 -0800

Registering passes automatically.

My colleague Adrian Sampson showed me a really good trick for running LLVM passes automatically, and luckily he wrote a really nice blog post about it.

LLVM Command line options

Thu, 30 Jan 2014 00:00:00 -0800

Command-line options

Okay, command-line flags are not the most exiting topic ever. But, potentially pretty useful when you're making a complicated pass and want to experiment with a bunch of different options. And I just have to say, this is some absolutely stunning C++ magic. I thought how LLVM smoothly handles strings with StringRef/Twine was impressive, but this is on a whole new level. But we'll start with the mundanes of actually using it in the simplest way possible.

Building LLVM on OSX

Wed, 29 Jan 2014 00:00:00 -0800

I've had a lot of headaches building LLVM from source to suit my purposes. It's fairly straightforward to download a source distribution and get something built, but I've had to refine how I build LLVM so that things are setup to be useful for my workflow. Namely, I need: