Root Access

5:00 PM

@HackToHell oh, another thing to keep in mind; while you're doing your testing, you will want to get some software that injects itself at the network layer on your test client, and randomly decides to drop packets and increase latency, so you can observe how your software copes with adverse network conditions

a company called Shunra sells proprietary software for this purpose, but if you need FOSS, I think there's a Linux kernel module that does it.

Bob

@allquixotic or literally run it on a laptop on GSM travelling through the countryside :P

allquixotic

@Bob you use less fuel by modprobeing the kernel module ;p

HackToHell

@allquixotic I was thinking an android phone for now until we can find a lower cost hardware solution

Bob

106

Q: Simulate delayed and dropped packets on Linux

I would like to simulate packet delay and loss for UDP and TCP on Linux to measure the performance of an application. Is there a simple way to do this?

HackToHell

@allquixotic ooh Okay

And I'm thinking of writing the backend using Go and mysql

allquixotic

5:02 PM

@HackToHell looks like the best answer to that question Bob just posted will work great on Android (probably, you may have to recompile the kernel)

HackToHell

Though I am a bit scared of mysql

allquixotic

130

A: Simulate delayed and dropped packets on Linux

netem leverages functionality already built into Linux and userspace utilities to simulate networks. This is actually what Mark's answer refers to, by a different name. The examples on their homepage already show how you can achieve what you've asked for: Examples Emulating wide area n...

@HackToHell for 1k requests per second, stay the flying fuck away from MySQL

3

just don't

use Postgres

hell, use a KV non-SQL database; use anything; just don't use MySQL

HackToHell

Why would I even need a relational database anyway.

Bob

10

Q: How do I simulate packet loss on an Android device?

I'd like to simulate packet loss on a rooted Android device to see how different levels affect video streaming quality. I've tried netem and iptables, without success: netem: # tc qdisc change dev rmnet0 root netem loss 0.1% tc qdisc change dev rmnet0 root netem loss 0.1% Android does not suppo...

2.3-gingerbread networking streaming

allquixotic

@HackToHell all you really need is KV

Key = ID; Value = double[] or float[]

Bob

5:04 PM

@allquixotic how well do current KV implementations scale, anyway?

I've never looked into that

HackToHell

looks up KV

Bob

most major RDBMSes have been tested under extremely high loads and they stand up well

allquixotic

@Bob extremely well, as long as the write operations you're performing on them are fairly simple... in this case, the process of storing the data is trivial in HackToHell's planned project, and in general DBs only ever struggle with write perf; read perf is generally great

HackToHell

Holycrap there are a lot of no-sql databses

allquixotic

a KV database would suck if you had to update the value of many keys very frequently, depending on values changed in other keys, etc

but in this case each key is literally independent and can be updated in parallel with no locking

and the read is just, well, read... it'll be lightning fast, not a bottleneck

Bob

5:06 PM

@allquixotic it'll be painful when time comes to query the data, though

unless you want to convert to a different format first

if you want real-time queries, KV probably isn't what you want (e.g. GPS bus-trackers)

allquixotic

@Bob generating statistics will probably have to involve something like pulling a static copy of the entire DB into RAM (if there's enough RAM) or streaming it out to JSON or XML on disk (if there's not enough RAM), then either processing it a la DOM or a la streaming (depending on the size of the dataset) on a separate server, so as not to disturb the DB server

I'm going to concede that Postgres and an optimized SQL statement would probably result in faster stats generation :P

but for just getting the shit in the database quickly, KV likely wins

HackToHell

I could push the database into a KV and then slowly(?) add it to a postgresql server ?

Bob

@allquixotic Hm. I'd test that, but I really can't be bothered doing anything right now.

HackToHell

So that it can be queried later and such

@Bob I am going to :D

As soon as I get a hang of this posgresql anyway

Phew so the backend is Go+Postgresql

Time to write some codez

Bob

@allquixotic I mean, if you want pure write speed you could just append to a file... just hope you never need to read it! :P

allquixotic

5:13 PM

@Bob at least a KV database will provide O(logish) lookup time for a specific key

traversing the entire dataset is probably going to be slow, though

one thing you could efficiently do with a KV is provide an individual user, upon request, a webpage displaying their stats for their specific vehicle

look up their key (super fast whee) --> pull out the values --> process --> render page

no problem there. the problem comes when you need to generate stats across ALL the vehicles

like SELECT * FROM stats WHERE speed > 120;

doing that in Postgres would be fast, but painfully slow (without some kind of in-memory data structure) in the KV

Bob

@allquixotic but if you need to store a history of positions...

that's actually relational data

you could store it with a key of id_timestamp, but then you'd be performing an O(n) traversal to find all possible timestamps anyway

allquixotic

oh... with the history you probably need to record time, too

HackToHell

fuck forgot about that

allquixotic

now, in an RDBMS, you could have a composite key of id and timestamp

HackToHell

The time too is needed :|

allquixotic

5:17 PM

the more I think about it, the more I feel that having a relational database server with 172 GB of RAM is going to be your (ultimate / production-grade) ideal backend

Bob

@allquixotic and it'd be easy to select * from stats where id = 123 order by timestamp

allquixotic

Postgres would work

Bob

or most recent 10, or where timestamp within last day, etc.

allquixotic

@Bob yeah, and since DBs' storage format and algorithms are extremely finely tuned to do stuff exactly like that as fast as possible, you don't have to reinvent the wheel by figuring out how to do that after you extract data from the KV

HackToHell

Can the database be split amongst different computers ?

allquixotic

5:19 PM

only thing you have to bear in mind is that, IF the statistics-gathering load becomes too severe, you might want to either have a replicated database, or take periodic static copies of the master DB into a separate stats DB, to save the CPU of the poor production DB

HackToHell

I could split the data based on Bus Ids

allquixotic

in other words, if tons of users like to hit your website and look up their stats and query stuff to and fro, that's going to use up CPU that your primary DB server should be using for storing data

so a static copy or replication would help

HackToHell

Oh boy, building a scalable app is no easy thing

Bob

incidentally, at 1k/second, you're looking at ~2.5 GB of writes per day: this is a link dammit markdown

that's... not much at all

allquixotic

!! s/markdown/commonmark/

ChatBot John Cavil

5:21 PM

@allquixotic incidentally, at 1k/second, you're looking at ~2.5 GB of writes per day: this is a link dammit commonmark (source)

allquixotic

:D

Bob

(that's assuming a 128-bit GUID, and 32-bit numbers for location and accel)

allquixotic

@HackToHell if it were easy, they wouldn't pay people at Amazon, GOOG, etc. high values of 6 figs salary to think of this stuff

Bob

(add 32 bits for timestamp, bringing it up a little bit, and add a bit for indexing etc)

allquixotic

@Bob even if you increased that by a factor of 10, an SSD and a commodity Ivy Bridge Xeon should be able to handle that, no?

Bob

5:22 PM

@allquixotic seems like it

HackToHell

@allquixotic That's pretty decent

code.google.com/p/protobuf

Google moves it from Google Code to github

allquixotic

there's a difference between designing for scalability and overengineering though -- to start out with, when you have 0 customers, you SHOULD really be running everything on a single, smallish server or VPS... only AFTER you start to get significant customer demand should you scale... your purpose in designing for scalability is to be ready to add servers or upgrade your server once the demand comes

Bob

in other words, good abstraction

allquixotic

and hopefully, scaling up won't require you to write any code, just tweak a few configuration files

Bob

you could even start off with just writing to sqlite

allquixotic

5:25 PM

yeah, for sure

HackToHell

Thing is it's just a Proof Of Concept kinda thingy

allquixotic

your code's use of threading or multiprocessing is going to play a big role in the performance though, so always think in terms of how you're going to deal with many simultaneous requests

Bob

with a scheme as simple as yours, it really doesn't matter how you store the data just yet

allquixotic

you're a Proof of Concept!

HackToHell

And it's better if I show them it works and it can be scaled

Bob

5:26 PM

make sure it's easy to change your code to use a different storage mechanism, and that's all

@allquixotic eh, depends

HackToHell

@allquixotic C++ and OpenMP ftw ? :D

Bob

the ideal model is more-or-less how current HTTP servers work

single thread listening on a socket, passing input off to worker threads that update the DB

but you don't have to start off with multiple threads - make it work first with a single thread! (just make sure you separate your DB, processing and listening code and it should be easy to split it up when the time comes)

HackToHell

Okay, I'll start with a simple implementation and then add the code for handling a lot of users

allquixotic

@Bob nginx does that, but it spawns several separate processes each with their own listening thread

doesn't it?

!! s/add/rewrite/

ChatBot John Cavil

@allquixotic Okay, I'll start with a simple implementation and then rewrite the code for handling a lot of users (source)

allquixotic

5:32 PM

just being realistic

HackToHell

uh huh

And I'm going to go with Go

Cause apparently it's fast and all

Bob

@allquixotic eh, I'm not too sure about the implementation details

nor why that would be a good idea :\

HackToHell

> func add(x int, y int) int {
return x + y
}

return type comes later in Go

Why would they do that o_0

allquixotic

@HackToHell func

Bob

> The only reason nginx uses multiple processes, is to make full use of multi-core, multi-CPU and hyper-threading systems. Even with SMP support, the kernel cannot schedule a single thread of execution over multiple CPUs. It requires at least one process or thread per logical CPU.

...I still don't see why it needs processes.

allquixotic

5:38 PM

play that funcy music white boay!

Bob

@HackToHell well you chose the language :P

strangely, I'd actually seriously consider nodejs for this

it seems quite popular these days :P

(ffsync, ghost, ...)

hm

HackToHell

@allquixotic I don't get it o0

allquixotic

nodejs also has a much better selection of libraries

HackToHell

@Bob so is Go :D

allquixotic

almost on the scale of ruby :D

Bob

5:39 PM

now I suddenly realise I have no idea what language I'd use for a similar project

C# is usually my go-to, but it's not quite as easy on *nix

also probably not great when you're looking at processing 1k r/s :S

hm. now I want to test that.

HackToHell

@Bob Start with Java :D

Bob

@HackToHell nothanks

allquixotic

@Bob I'm sure v8 could handle it ;p

HackToHell

@Bob hehe

allquixotic

I wonder how much faster my Ruby program could run on NodeJS? or if it would be faster at all?

Bob

5:43 PM

I like my TAP

allquixotic

I'm using MRI 1.9.3. the bottleneck seems to be in running Tarjan's algorithm over the directed graphset every time I add a node.

Bob

(used the wrong abbriev... dammit)

HackToHell

tour.golang.org/#10

Why would they even do that

Debugging Go code must be hell

I am too used to super strict Java :|

Bob

> the difference between node and async c# is 10-15% either way depending upon what you're doing

allquixotic

I need help thinking about this algorithm...

Bob

5:48 PM

hm

@allquixotic quack?

allquixotic

:D

yes. thanks.

my problem: on a dataset of ~7000 nodes of a polytree (connected digraph), whose keys are represented by an integer, my task is to ensure that a cycle is never introduced into the digraph -- firstly because it causes infinite loops in arbitrary parts of the graph API, and secondly because I need to detect cycles in the data and report them as problems, and remove the minimum number of nodes to restore the rest of the graph as non-cyclic.

to do this, currently, I use Tarjan's algorithm to sort the entire polytree each time I add a node, before accessing any properties of it; the algorithm throws an exception if a cycle is found

the problem is that the sorting and the cycle detection go hand in hand with this algorithm, and you can't really have one without the other; furthermore, if I don't try and sort the entire polytree each time I add a node, I could hit one of those infinite loop conditions

so I can't wait until after I've compiled the entire tree and then do a sort, because I have to access properties of the nodes during tree construction

so it's the O(Tarjan's Algorithm * n) -- it's the * n part that is killing me

it's taking 25 minutes to run (some of that time is spent parsing XML, but a lot of it is being spent sorting the polytree)

Bob

@allquixotic could you detect a loop while trying to access said properties?

allquixotic

@Bob the only reliable way I know of to do that would be to run Tarjan's algorithm before each property access, which will be even slower

based on the complexity of the graph, I can't look at any specific number of iterations of the internal loops of the properties (many of which attempt to traverse the entire tree, which is when they get caught in a cycle) and say "aha! it's looped too many times, so that must be a cycle!"

Bob

@allquixotic but wouldn't it be a cycle if it encounters the same node more than once?

allquixotic

it could be acyclic but just take a really complicated route to traverse

Bob

5:56 PM

so "too many times" => n > 1

allquixotic

hm...

so you'd have to keep a list of every node you've traversed during property access (preferably a hash, for lookup performance), and each time you encounter a new node, add it to the hash if it's not there, and if it's already there, throw an exception

that sounds pretty algorithmically complicated, though easy to implement :S

Bob

unless some condition has changed; if you start a deterministic traversal from one state and end up at that state again, you're in a cycle

(where a state includes current node and anything that might affect choice of descendants to traverse)

allquixotic

yeah, but the problem is detecting that isn't very easy or fast, and I'm trying to avoid paying the cost of the cycle detection so often, but it seems that in attempting to avoid Tarjan's I end up reinventing a cycle detection algorithm of my own :P

instead it seems I'd be writing the naive cycle detection algorithm (as I described above)

Bob

@allquixotic well, how often does a cycle occur?

if it's frequent, then checking with Tarjan's every time might be more efficient

allquixotic

@Bob it's a bit of an edge case. but, OTOH, if it's not detected, it can hang the program and make it nearly impossible to diagnose.

Bob

6:00 PM

if it's rare, then it might be more efficient to assume no cycles (no explicit Tarjan's, just a check-as-you-traverse) and simply check once at the end

@allquixotic the goal of cycle detection during traversal would be to prevent a non-terminating traversal, not detect the existence of any cycle in the graph

allquixotic

it's pretty rare in the test data that I have, but if I don't check, I'll be maintaining hashes and doing repeated additions and lookups in them as I traverse and add, which will be almost as bad as doing Tarjan's each time (or worse, if Tarjan's is faster than that)

Bob

if your traversal does not cover all possible paths, then it could be more efficient than Tarjan's (which guarantees no cycles in the entire graph)

@allquixotic but is Tarjan's faster than that?

I would've thought that to be pretty cheap

especially if the paths you cover during your property access is a small subset of all paths

allquixotic

the complexity of Tarjan's is on the order of linear with the number of edges and nodes of the graph

Bob

of course, you lose the short-circuiting behaviour of an explicit Tarjan's check since you might miss a cycle until the final check

@allquixotic So... did you ever profile just how much of that 25 mins it took?

And what's the constant? (linear... and? is the constant considerably greater than maintaining a hash table? which will be O(no. of property accesses * path length))

allquixotic

@Bob no, but that entire 25 minutes was consumed by in-memory operations (no disk, no network), and the process wasn't swapped out (no memory pressure), so it would have been either that, or Ruby's hashmap methods being slow... I think the XML was already processed before then too

Bob

6:06 PM

@allquixotic But how much time did Tarjan's algo take compared to time taken for property accesses, adding nodes, etc?

allquixotic

good point; I should probably measure that with ruby-prof

Bob

If it's insignificant, then you're looking in the wrong place.

O_O owin.org

allquixotic

@Bob profiling now :D

Bob

haneycodes.net/to-node-js-or-not-to-node-js

I... never knew about this.

This is awesome! :D

And it works on mono! :D

@allquixotic ^

dumping IIS => fast C#/.NET webserver

allquixotic

6:22 PM

@Bob Yeah, I'm reading the article now. It looks nice.

Bob

@allquixotic the guy who wrote it apparently got a job at SE shortly after that post O_O

about two months after

allquixotic

@Bob lol nice

DemCodeLines

So speedtest.net shows my download speed as "10 Mbps", but the Windows 10 Technical Preview ISO is downloading at 855Kbps...why?

Bob

is that 855Kbps or 855KB/s?

!! s/K/k/g

ChatBot John Cavil

@Bob is that 855kbps or 855kB/s? (source)

Bob

6:37 PM

@DemCodeLines ^

DemCodeLines

KB/sec

That Brazilian Guy

@DemCodeLines Because it is working as intended.

DemCodeLines

Then what does the download speed mean?

That Brazilian Guy

Hm.... Data throughtput into your computer?

speedtest measures in megabits per second, downloaders usually measure in kilobytes per second (or megabytes)

Bob

37

A: What is the difference between a kibibyte, a kilobit, and a kilobyte?

1 KiB (Kibibyte) = 1,024 B (Bytes) (2^10 Bytes) 1 kb (Kilobit) = 125 B (Bytes) (10^3 Bits ÷ (8 bits / byte) = 125 B) 1 kB (Kilobyte) = 1,000 B (Bytes) (10^3 Bytes) It's the same way with any SI prefix; k (1x103), M (1x106), G (1x109), so, by extension: 1 MiB (Mebibyte) = 1,048,576 B (Byte...

simply: 8 bits in a (modern/x86) byte

That Brazilian Guy

6:41 PM

Yeah, that

Bob

@allquixotic well... good luck with your graph

I'm gonna try to get some sleep :P

allquixotic

@Bob aww :(

I profiled it!

the threads that spend all their time sleeping can effectively be discarded as not relevant

aaaand I spend 61.51% of my time doing Array.concat -_-

Bob

@allquixotic fine, fine, I'll come back :P

allquixotic

<3

Bob

I have no idea how to read that O_O

allquixotic

6:53 PM

all the busy work happens in Thread ID: 30154800

%self is the percentage of total CPU time of the program (across all threads) that was spent, BUSY, on that method

%self      total      self      wait     child     calls  name
61.51   1036.495  1036.495     0.000     0.000 26098607   Array#concat

that sums it up, basically

Bob

gimme a sec gonna run it through a couple filters

allquixotic

I've determined that you can ignore all the data except for Thread ID: 30154800 (do a find for that)

and only the top 10 or so entries from that list are even worth considering

 %self      total      self      wait     child     calls  name
 61.51   1036.495  1036.495     0.000     0.000 26098607   Array#concat

	called from:  Tree::TreeNode#each (C:/MyCode/runtime/lib/ruby/gems/1.9.1/gems/rubytree-0.9.4/lib/tree.rb:522)

  3.05   1543.689    51.338     0.109  1492.242     9247   Tree::TreeNode#each
  C:/MyCode/runtime/lib/ruby/gems/1.9.1/gems/rubytree-0.9.4/lib/tree.rb:522
	called from:  Tree::TreeNode#tsort_each_node (C:/MyCode/tools/MyCode/RestApi.rb:14)  MyCode::RestApi#getSubfolders (C:/MyCode/tools/MyCode/RestApi.rb:399)

there, that's all that's really needed from that ~250kb log

Bob

@allquixotic So where's the other 30%?

allquixotic

@Bob other threads, sleeping, little 0.01%s here and there (hundreds of them), etc

Bob

Though, that's mostly insignificant compared to the Array.concat...

allquixotic

6:58 PM

    # Traverses each node (including this node) of the (sub)tree rooted at this node
    # by yielding the nodes to the specified block.
    #
    # The traversal is *depth-first* and from *left-to-right* in pre-ordered sequence.
    #
    # @yieldparam node [Tree::TreeNode] Each node.
    #
    # @see #preordered_each
    # @see #breadth_each
    #
    # @return [Tree::TreeNode] this node, if a block if given
    # @return [Enumerator] an enumerator on this tree, if a block is *not* given
    def each(&block)             # :yields: node

that code, specifically, the 5th or 6th line from the bottom (depending on if you count whitespace), calls concat so many times that it literally is accounting for 61% of the CPU time of the entire process

and that's with a very large amount of XML being processed in the same process space, network I/O, etc

in fact, the XML stuff is so fast that it's in the noise

Bob

@allquixotic And what calls this function?

Can you minimise those?

allquixotic

@Bob crap. Tarjan's algorithm calls it.

Tarjan's is a mixin that implements itself partially by calling each

Bob

Welp.

(I was kinda expecting that)

allquixotic

it would appear that Tarjan's itself may not be all that painfully intensive, but asking the rubytree to each its nodes is painful

come to think of it, that code block I just pasted looks pretty stupid, doesn't it?

I mean, why do you need to dynamically allocate an array and concatenate a bunch of nodes onto it just to get a stack of nodes based on the children of the current node?

if this were C, you'd use pointers. bam, done.

Bob

shrug

allquixotic

7:04 PM

(this isn't my library; you can view the code online here)

Bob

a LL would be better here

allquixotic

@Bob not sure you can represent a digraph with a linked list

interestingly, the clone that method calls is eating up about 1% of the CPU, too. well, that makes sense, since clone is memory allocation

Bob

@allquixotic node_stack

you'd avoid having to resize (and possibly copy) the array

allquixotic

hmmmmmmmm you're right

Bob

or maybe if you can reserve a larger capacity?

allquixotic

7:07 PM

will have to look at the Ruby impl of Array.concat too

Bob

would it hurt to allocate an array with the max number of elements in the graph?

6

A: Ruby Array concat versus + speed?

The answer lies in Ruby's underlying C implementation of the + operator and the concat methods. Array#+ rb_ary_plus(VALUE x, VALUE y) { VALUE z; long len, xlen, ylen; y = to_ary(y); xlen = RARRAY_LEN(x); ylen = RARRAY_LEN(y); len = xlen + ylen; z = rb_ary_new2(len);...

@allquixotic You might also save a bit with array.push, but that's seriously microoptimising where LL or reserving capacity is likely better

allquixotic

wait, if all concat is doing is splicing two arrays together by making a pointer from one to the next, isn't it almost like a linked list?

Bob

@allquixotic Hm. The pseudocode of the latter link says otherwise

determine new length as ary.length + other_ary.length
if new length makes ary over capacity
  increase ary capacity
copy other_ary into ary

allquixotic

ahhhh

Bob

1450 static void
1451 rb_ary_splice(VALUE ary, long beg, long len, VALUE rpl)
1452 {
1453     long rlen;
1454
1455     if (len < 0) rb_raise(rb_eIndexError, "negative length (%ld)", len);
1456     if (beg < 0) {
1457         beg += RARRAY_LEN(ary);
1458         if (beg < 0) {
1459             rb_raise(rb_eIndexError, "index %ld too small for array; minimum: %ld",
1460                      beg - RARRAY_LEN(ary), -RARRAY_LEN(ary));
1461         }
1462     }
1463     if (RARRAY_LEN(ary) < len || RARRAY_LEN(ary) < beg + len) {

allquixotic

7:13 PM

looks like some memory copying to me

it seems to work vaguely similar to Java's ArrayList.Add

Bob

@allquixotic Looks like the behaviour of rb_array_splice is to resize and push, not link.

ary_double_capa(ary, alen);

@allquixotic reserving capacity is probably the easiest course of action

> An array can also be created by explicitly calling ::new with zero, one (the initial size of the Array) or two arguments (the initial size and a default object).

You'll still have some copying going on, but that can't be avoided without modifying the implementation of node.children

allquixotic

well, even though Kernel.clone takes up 1.59% of the time, I honestly don't think it's worth trying to not clone shit... I'll continue letting it clone when it wants to; if I can improve the perf of the Array.concat by using a list instead, that would be huge

Bob

Hm, actually... you can avoid .clone by directly reading node.children and turning that into a LL

not sure if it'll be faster or slower than reserving capacity

certainly more difficult

allquixotic

I think Array.unshift creates a linked list

Bob

Once you learn that shift/unshift are like push/pop on the other end of the array, you can mentally drop the 'f' from the name of the methods to remember which one 'dumps' elements and which one 'inserts' them. :) — Phrogz Jan 21 '11 at 17:37

lol

allquixotic

7:18 PM

node_stack.unshift(current.children.clone) ?

Bob

@allquixotic apparently not

I cbf digging into the actual impl code

Ugh. Looks like the 'initial size' crap doesn't work as hoped

there doesn't seem to be a direct equivalent to .NET's .Capacity

pity. that would've solved this pretty well

@allquixotic looks like you might be stuck iterating over current.children and converting it to a LL, if you want to go that path

I'm not sure how much faster that would be, if any

on one hand, you're avoiding a lot of resize + copy (assuming node_stack grows quite big)

on the other, you're replacing C resize + copy with ruby loops... the latter likely won't be as well optimised

allquixotic

@Bob interestingly, the node_stack is never returned to the caller; it's simply used to keep track of where this code is in the stack

the line yield current basically calls an anonymous method ("block") provided by the caller, passing current as a parameter

Bob

@allquixotic hm? pretty standard, isn't it?

allquixotic

that's what the caller cares about

the caller doesn't care how we keep track of the child nodes internally

Bob

it's basically a translation of recursion into a loop with a stack variable

the problem is ruby doesn't give you any easy way to prevent excessive resizing, and no native LL implementation either

@allquixotic Though, if this is mostly called through Tarjan's, and accounts for 60% of CPU usage, would the earlier suggestion of checking for cycles during traversal be better?

allquixotic

7:26 PM

maybe -_-

Bob

Would the use of hash tables there account for >50% of current time taken? If not, then that's a better way to go.

Again, you lose short-circuiting behaviour of repeated Tarjan's (if there's an unnoticed path, it'll only be detected by the final check at the end), but since that's supposedly rare then it should be better overall

allquixotic

hmmm

Bob

I suppose the next thing to do would be to compare LL here against avoiding Tarjan's.

The former should be better than resizing arrays, but replacing native code with interpreted code might take make it worse. The latter... pretty much test and see? :P

allquixotic

yeah

thanks :P you should sleep now

Bob

lol

good luck

Oliver Salzburg

7:57 PM

Anyone else on this already?

tombull89

Adding it to my WDS server now.

That Brazilian Guy

@OliverSalzburg Windows 10? O.o

Oliver Salzburg

@ThatBrazilianGuy Yeah. preview.windows.com

That Brazilian Guy

But... What about Windows 9?

Did they just skip?

History of Microsoft Windows

In 1983, Microsoft announced the development of Windows, a graphical user interface (GUI) for its own operating system (MS-DOS). The product line has changed from a GUI product to a modern operating system over two families of design, each with its own codebase and default file system. The 3.x and 4.x family includes Windows 3.0, Windows 3.1x, Windows 95, Windows 98, and Windows ME. Windows for Workgroups 3.11 added 32-bit networking and 32-bit disk access. Windows 95 added additional 32-bit capabilities (however, MS-DOS, some of the kernel, and supplementary utilities such as Disk Defragme...

They did.

Oliver Salzburg

@ThatBrazilianGuy They "explained" that in the introduction. It's not an incrementally improving product, it's so mindblowingly revolutionary, that it's 2 versions better

allquixotic

8:16 PM

@OliverSalzburg so, Java 5? :P

from Java 1.4 to Java 5 :P

Oliver Salzburg

I like it so far

...he says after 5 minutes of use

tombull89

8:32 PM

IT'S ALIIIIIIIIIIVE

Liking it, liking it.

the Metro apps don't open full screen :')

Mokubai

Still waiting on my download. Damn slow interwebz.

tombull89

"use the start menu rather than the start screen" oh niiiiiiiiiice

Mokubai

At about 7:30 PM Firefox told me it was going to take 1 hour 45 minutes to download. It still has 1GB to go and will take another "40 minutes"

Psycogeek

8:54 PM

@tombull89 the return of windows ? what a concept

Tom Wijsman

9:07 PM

@OliverSalzburg Yes, but I boot from VHD. :D

Mokubai

@TomWijsman Forgot you could do that, any link to a guide?

Tom Wijsman

@Mokubai: Only step 1 and 2 of technet.microsoft.com/en-us/library/hh825691.aspx and then bcdboot V:\Windows

Fairly easy to accomplish; create and format VHD, apply image, make it available in BCD.

Mokubai

Awesome, cheers :)

Tom Wijsman

You can change the default or lower timeout of the boot menu in msconfig.

Mokubai

Was going to put it in a VM, but VHD would probably be better...

Tom Wijsman

9:16 PM

Did the same when Windows 8 came out; I was almost about to not try Windows 10, until I remembered Boot from VHD.

Not sure if all Windows 8(.1) versions can do Boot from VHD; IIRC, it was introduced as a professional feature in Windows 7.

Mokubai

I have Win8 Pro (actually paid for it) so I should be fine.

DemCodeLines

Y'all still downloading Win10?

yesterday, by DemCodeLines

"Because 7 ate(8) 9"... sigh

This is what it says on Windows feedback page:

> And don’t worry about hurting our feelings—be honest!

lol

Stack Exchange

9:50 PM

0

Q: Flag as off topic - move to linux/unix community is missing

I have tried to flag this question as off-topic, to move it into Unix & Linux community. However there is no option for that (or I don't see it?): Is this missing on purpose? I know there is Linux tag in here, but I would move that anyway, because people probably can get better Linux support...

feature-request off-topic

Psycogeek

10:34 PM

"I just assumed that the EXE is a download manager for the ISOs, like I've seen in the past. Next thing I know my CURRENT machine is rebooting to update to Windows 10!" answers.microsoft.com/en-us/windows/forum/… Is it wrong to laugh :-)

Psycogeek

10:50 PM

I should test the thing and provide feedback, so Ignoble Detractors Ignoring Obvious Trouble Spread do not set my destiny.