rbranson

14 ms·

121.
by rbranson 7y ago
By default Django closes PostgreSQL connections after handling each web request. I’m not certain but what we’re probably seeing here in latency is mostly the substantial overhead in setting up a PG connection. Redis will still be faster but
122.
by rbranson 7y ago
40,000 people per day? Seems a little high.
123.
Western Digital's Ultrastar DC HC330: 10 TB HDD with 40% Faster Writes(anandtech.com)
2 points by rbranson 7y ago|0 comments
124.
by rbranson 7y ago
The order-of-operations is that it checks RocksDB first, writes to Kafka, and then writes to RocksDB. If the write to Kafka fails, it re-positions itself in the input topic stream based on the offset annotation in the output topic's la
125.
by rbranson 7y ago
A counterpoint is our ingest-time deduplication system at Segment: https://segment.com/blog/exactly-once-delivery/ It's done at ingest time because Segment has a completely different use case. Message data fa
126.
by rbranson 7y ago
Citus is a completely different animal. This is just vanilla PostgreSQL with a different storage backend. No PostGIS either because just like RDS, no custom native extensions.
127.
Serving 100µs Reads with 100% Availability(segment.com)
2 points by rbranson 7y ago|0 comments
128.
by rbranson 7y ago
Someone is going to have to put up the money to buy the fleet of autonomous vehicles.
129.
by rbranson 8y ago
This could essentially be said about any corporation.
130.
by rbranson 8y ago
I've always wondered what the best concurrency approach would be for a fully compatible fully concurrent Redis. The single-threaded nature makes it naturally expose some strict semantics to the user, which is tricky. What are you plann
131.
by rbranson 8y ago
According to the README all access to the data is effectively single threaded. Only the network work is actually multi-threaded. > Access to the core hash table is guarded by spinlock. Because the hashtable access is extremely fast this
132.
by rbranson 8y ago
This seems a little hard to believe. Do you have a source?
133.
Scaling up reporting on high-cardinality metrics(segment.com)
6 points by rbranson 8y ago|0 comments
134.
V100 Server On-Prem vs. AWS P3 Instance Cost Comparison(lambdalabs.com)
141 points by rbranson 8y ago|127 comments
135.
What Are Developer Community Projects?(developermode.com)
1 points by rbranson 8y ago|0 comments
136.
by rbranson 8y ago
I don’t think the author really addresses the examples presented. The example of the FAST decision tree is based on a code generator, an abstraction which is presumably more elegant than the generated code which is littered with goto statem
137.
by rbranson 8y ago
Certainly one way of categorizing western society is white supremacy and dominance of white culture.
138.
by rbranson 8y ago
These instances already deliver better cost efficiency. The a1 instances are within 5% of the performance of similar c5 instances, but are 40% (!!!) cheaper.
139.
by rbranson 8y ago
We run an auto-scaled pool of C5 instances that host stateless containerized services. T* instances don’t make sense since we run them pretty warm all the time. If we need less capacity, we scale down. The appeal for this type of workload i
140.
by rbranson 8y ago
FoundationDB provides the building blocks. Secondary indexes can be built on top of the regular primary index. Transactions make it possible to maintain a consistent 2I using a K/V interface.
141.
by rbranson 8y ago
James Hamilton talked about their commitment to 25GbE hardware at Reinvent in 2016. Fast forward to ~23m. https://youtu.be/AyOAjFNPAbA
142.
by rbranson 8y ago
It has to do with hardware support for virtualized networking. Instances capable of SR-IOV or with the ENA can sustain millions of PPS. It makes a huge difference.
143.
by rbranson 8y ago
The (undocumented) PPS limitation on EC2 instance types before they added SR-IOV NICs is around 150K PPS. If you had your own full machine — usually the top size of a given instance class, but no guarantee — this would be pretty consistent.
144.
by rbranson 8y ago
Maybe it’s supposed to be ironic given the title.
145.
Driving Adoption of a Design System(segment.com)
5 points by rbranson 8y ago|0 comments
146.
We test in production. You should too(segment.com)
2 points by rbranson 8y ago|0 comments
147.
by rbranson 8y ago
Maybe I’m missing something, but the way this reads seems to attribute Spanner’s concurrency control entirely to TrueTime. Spanner still uses partitioned Paxos to establish consensus and 2PC for multi-partition write transactions.
148.
by rbranson 8y ago
A company that is net losing money can't get a loan for 90% of their forward revenue.
149.
by rbranson 8y ago
This sounds really interesting. Can you expand on this installation?
150.
Buybacks are the wrong boogeyman(theirrelevantinvestor.com)
2 points by rbranson 8y ago|0 comments

More ›