donavanm

62 ms·

661.
by donavanm 12y ago
A key point that's glossed over is computational complexity on the senders side. The existing congestion & flow control mechanisms are quite simple. In contrast these "algorithms have more than 150 rules." Additional stat
662.
by donavanm 12y ago
So an RE failed and flipped to the secondary. They run mlag from the agg layer to their tors. Said mlag had a grey failure. The time to recovery was predominantly fault detection. After detection they upgraded the os on the known good switc
663.
by donavanm 12y ago
+1, in general. To adda bit of content: 1) Power Hungry, the total power envelope of a rack position is the limiting issue. You can fit 16-20 disks per RU no problem. The constraint on total density is your 5kVa or 10kVa power budget. Watt
664.
by donavanm 12y ago
Or lets say thats about $12,000 a year. Very conservatively thats 1/10th of a developers cost. Could you really build, and operate, a truly equivalent service with 1/10th of a developer? Im dubious. There are many cost/effici
665.
by donavanm 12y ago
Do you require dedicated IPs and a customer cert? If not CloudFront has free HTTPS SNI support and a free tier of 50GB & 2,000,000 HTTP(S) requests per month. So that would be your own domain name, your own ssl cert, and 2M requests per
666.
by donavanm 12y ago
Trust but verify, ie tamper resistant auditing, works for low sensitivity data. It's also an investment you'll never outgrow. When you get larger, or more sensitive, its time to implement something like a Two Man Rule. Take your p
667.
by donavanm 12y ago
Ha! I did have an outage because route announcements and data plane were on the same host a few years back. Having a separate health check service & route server is a trade off of complexity vs control. I could see the argument when y
668.
by donavanm 12y ago
This works well if you need to push high bit rates and are lookig for relatively simple load balancing. A trident box, ala juniper qfx, can push a few hundred gbs for ~$25,000. Thats an incredibly low price point compared to any other lb so
669.
by donavanm 13y ago
Thats how almost every modern tracing facility works. Headaches are generally simplified by no malloc, no backtracking, no preemption and a limited DSL with strict compilation.
670.
by donavanm 13y ago
Documentation is sparse in details. Is this built on top of kprobes[1]? If so a comparison with perf, systemtap, et al is probably more appropriate. Do you support utrace/uprobes or similar for userland work? Is data selection/man
671.
by donavanm 13y ago
I havent seen a point by point comparison. Feature set, and even the API, are basically identical for now.
672.
by donavanm 13y ago
Lets put some math on the network costs. An oft quoted cost is $1/mb at p95 for decent transit. 1 mbs * 2,592,000 seconds = 2,592,000 mb per month, / (8 * 1024 gigabytes) = 316GB transferred per month. Dividing our $1 / 316G
673.
by donavanm 13y ago
Amazon has deprecated & disabled old API versions, SOAP & security issues and the like. I think they removed some of the old SimpleDB eventually consistent Query calls & made strongly consistent Select. Some of the regional API
674.
by donavanm 13y ago
File a case with customer service when a package is late. Youre right that youve paid for two day shipping. And customer supports SOP is to extend your prime term for one month when an item is late. Besides that do you live in a rural area
675.
by donavanm 13y ago
He also wrote for The Inquirer, and before that The Register, and before that ... some other print trade mag I can't recall. This goes back to the early 2000s when those papers actually broke stories and not press release. IIRC Semiacc
676.
by donavanm 13y ago
Intrinsity did design, not process. They did some design work for Samsung around the A8/Snapdragon days. Shortly after that Apple acquired Intrinsity and brought them in house ala PA Semi.
677.
by donavanm 13y ago
Technically it has to be applied at the edge; effectively where the end users are aggregated. It's not really feasible for transport networks to do anything like this. The second is there's little/no business benefit to appl
678.
by donavanm 13y ago
If youre a Prime member and a package misses it's promised delivery date you should contact customer support. SOP is to increase your prime subscription by one month.
679.
by donavanm 13y ago
To clarify, systemtap needs UTRACE or UPROBES kernel support to trace user processes. Without those you can still inspect kernel functions. IIRC UTRACE was mainly supported by Redhat, and available intheir kernels for quite a few years. In
680.
by donavanm 13y ago
> The livery doesn't appear to be just for the Superbowl It's new. There's some youtube footage of the same aircraft doing test flights a few months back. Plain white, and weathered, fuselage and tail.
681.
by donavanm 13y ago
S3 keys use the leading ~16 bytes for partitioning, most significant first, as I recall. Prefixing with guids is ideal, epoch timestamps are pathologically bad, a "random" char should be enough for 99% of users.
682.
by donavanm 13y ago
I was thinking of solutions for 10s of gbs on todays x86 boxes. Dollars and power are both a budget, so it's all a tradeoff. WRT to the opteron A1100 yes, I could see your comment. Something like a box of A1100 blades plugging 802.3ap
683.
by donavanm 13y ago
came up a while back https://news.ycombinator.com/item?id=6632576
684.
by donavanm 13y ago
At $400 I assume youre thinking of something like an Intel X520 DA2 plus optics. If you can tolerate the power you can do 82599 with dual Phys for more like $150-200. Obviously I have no idea on the network controller or its sdk/driver
685.
by donavanm 13y ago
Optics are a big chunk also. A new controller with DAC Gbe phy is probably more like 7-8W. They only need to do the controller on this chip. A couple watts for the Phy are part of the motherboards budget.
686.
by donavanm 13y ago
A CNAME and an ALIAS both reference a specific label/name. A DNAME applies to a tree, and subcomponents. Wikipedia has a decent example of DNAME usage.
687.
by donavanm 13y ago
Yeah, around 1% of resolvers seriously disregarding TTLs is what Ive seen. But if thats the argument theres absolutely no difference in cache lifetime of a single A backed by an ALIAS resolution and the typical A + CNAME. In either case all
688.
by donavanm 13y ago
Can you illustrate this with an example? How about explicit records and a 60 second ttl on A and CNAME to humor me. If implementing recursion based ALIAS foo.mydomain RRs MUST respect the TTL of bar.otherdomains RRs. With that proviso Im mi
689.
by donavanm 13y ago
Im missing quite a lot in this article. Personally i have my reservations about ALIAS. Recursion backed implementations in particular are full of Dragons and sharp corner cases. It's a shame that its missing any substantial critisicm o
690.
by donavanm 13y ago
Instead of netstat(8) or ss(8) check out /proc/net/sockstat and /proc/net/netstat and /proc/net/tcp. Might as well save a fork and some context switches. net.ipv4.tcp_rmem=8192 873800 8388608

More ›