The Comms Room

Joel E Salas

00:00

github.com/chassing/linux-sysadmin-interview-questions

thoughts?

HopelessN00b

Personally, I think that's a failing of StackExchange, deleted the stupid/funny/retarded ideas. They should be migrated to an appropriate section, like "serverfault.com/funny" or "serverfault.com/epicfail" or whathaveyou.

@JoelESalas Not especially limited to Linux, but tha might be a good thing.

Joel E Salas

@HopelessN00b Yeah, I like the emphasis on network questions

I don't know if I could answer some of the string manipulation ones without using google etc

Andrew

> Describe the general file system hierarchy of a Linux system.
A mess.

HopelessN00b

@Andrew Yeah, but what OS is that not true of?

Andrew

@HopelessN00b an abacus?

ewwhite

00:05

@MatthewIfe Sorry... Chicago snow

HopelessN00b

@JoelESalas That's a problem with all interview question sets that get down to specifics. It doesn't matter to actual job performance whether or not the person can actually answer your specific question, just that they know how to get the answer, which 90%+ of the time is most quickly and easily done "from Google.")

Andrew

> No manual entry for tuc

Michael Hampton

@HopelessN00b I am pretty sure a screenshot was taken of it at some point, but I can't find it. And I don't (yet) have 10K on SU.

ewwhite

@MatthewIfe No alarm in screen

shorter delay

9:07:48.910152 read(0, "\33", 256)     = 1
19:07:48.910497 read(0, "", 255)        = 0
19:07:49.310351 read(14, "\20\357\1\0002013122811341382\0\0\0006\v\0Q\n\0\367\v9"..., 16384) = 4477
19:07:49.310727 read(14, "", 4477)      = 0
19:07:49.310869 write(1, "\33[6;17H\33[25;41muce Pro          "..., 1022) = 1022

Matthew Ife

I guess trying to get a pty in there is better.

Andrew

00:10

the problem with the ipv6 questions is that ipv6 isn't being taught.

mossy

^ can confirm

ewwhite

@MatthewIfe With screen, the full trace is:

) = 1 ([{fd=0, revents=POLLIN}])
19:09:57.309761 ioctl(0, SNDCTL_TMR_START or TCSETS, {B38400 -opost isig -icanon -echo ...}) = 0
19:09:57.309807 ioctl(0, SNDCTL_TMR_TIMEBASE or TCGETS, {B38400 -opost isig -icanon -echo ...}) = 0
19:09:57.309842 read(0, "\33", 256)     = 1
19:09:57.309873 ioctl(0, SNDCTL_TMR_START or TCSETS, {B38400 -opost isig -icanon -echo ...}) = 0
19:09:57.309901 ioctl(0, SNDCTL_TMR_TIMEBASE or TCGETS, {B38400 -opost isig -icanon -echo ...}) = 0
19:09:57.309928 read(0, "", 255)        = 0

HopelessN00b

Alright, here's a fun question to throw out there... what AD forest name would you use for "test" if some dumbfuck already used test.company.com as a subdomain of an AD forest named company.com?

Andrew

@HopelessN00b shoot them

Journeyman Geek

@MichaelHampton asleep, apparently ;p

HopelessN00b

00:12

@Andrew Yeah, well he's gone (and out of rifle range), but the problem, and question remain.

Michael Hampton

@JourneymanGeek Heh. Can you expand all the comments and screenshot this for our amusement? :)

Journeyman Geek

@MichaelHampton: but of course

Michael Hampton

@JourneymanGeek THanks :)

HopelessN00b

Gotta deploy a SharePoint farm into "test" tonight, and the prospect of deploying it to the existing test forest is... unpleasant. I'm frankly wondering how much of a shitstorm I'd create for myself if I just stand up a new test.company.com forest. <sigh>

Andrew

@HopelessN00b is there a "test" department?

like, R&D?

Katherine Villyard

00:14

@HopelessN00b noreallyimeanittest.company.com. ;)

HopelessN00b

@Andrew Well, we have a QA department for "testing" the food products we make... but the test subdomain is used for "test" IT systems, not testing salad dressing.

Joel E Salas

@HopelessN00b What's the generally accepted mechanism for test SP environments?

Journeyman Geek

5

Michael Hampton

@MDMoore313 See above

HopelessN00b

@KatherineVillyard I might just end up using "testing.company.com" and be done with it. LEt it be someone else's problem in a few months here.

Journeyman Geek

00:17

(The joys, and the sheer horror of firefox's dev console)

HopelessN00b

@JoelESalas Deploying it into a test environment, I think. Problem being our test environment is a domain in that damn misnamed forest of company.com. Grrr.

Joel E Salas

@HopelessN00b You can frontload the effort and do it right, or you can add technical debt and do it later

Andrew

@HopelessN00b "If you or your client really is a spammer, do let me know" - classic

mossy

million contacts lol

HopelessN00b

@JoelESalas Yeah, not gonna happen. It was a hell of a fight to even get them to let me stand up a proper corp.company.com forest. They're not going to be willing to wait until we make the company.com and test.company.com domains go bye-bye. :(

Michael Hampton

00:21

@JourneymanGeek I swear there are comments missing... I may have to become a SU mod just so I can read the deleted comments.

Joel E Salas

@HopelessN00b I'm generally able to do 70/30 do it right/add debt at hautelook

HopelessN00b

@JoelESalas Yeah, one of the many criteria I'm looking at for $[nextjob]. Getting depressing... damn depressing... to see just how many companies do it wrong, and that they generally pay lots more than the companies that want to do it right.

Joel E Salas

@HopelessN00b Good thing we work for the sake of money and not for the fuck of it

more to do is always better, and more money to boot??

HopelessN00b

@JoelESalas Enh. For me that seems to have translated into more paperwork, more people I want to kill and exponentially less to do with technical ability. So, I'm seriously considering less/same money somewhere that does it right... if I can ever find such a place, of course.

Matthew Ife

In the middle of tryinbg a pty

Journeyman Geek

00:26

@MichaelHampton: can't help there... for now ;p

Joel E Salas

@HopelessN00b My way also has more options :D

ewwhite

@MatthewIfe oh my!

HopelessN00b

@JoelESalas I dunno. "More of the same shit" doesn't seem like the same as "more options" to me.

Joel E Salas

@HopelessN00b Very true

Mark Henderson

So, quick nerd question: I have a language that only has basic math (+-/*^%) and I need to bitwise operations. Is this possible? Mainly, I need to get the binary representation of a smallint

Matthew Ife

00:31

@ewwhite probably not straight-forward btw :-\

Mark Henderson

Once I can get the value into individual bits, I'm good from there on

HopelessN00b

@MarkHenderson log()?

Andrew

@MarkHenderson divide by 2, check remainders

Mark Henderson

@HopelessN00b Yep have log

Andrew

@MarkHenderson log(number) / log( 2 )

HopelessN00b

00:32

Yup, what Andrew said. That's how you convert bases.

Andrew

Wait. You're not asking a programming question, are you?

Mark Henderson

@Andrew Awesome

@Andrew yeah I'm not a full time sysadmin

It's just the part of my job I like the best

Andrew

@MarkHenderson it's not SQL is it?

or is it something estoric / evil?

Mark Henderson

@Andrew Nah it's not. SQL Server has bitwise operations

Joel E Salas

@Andrew It's probably some shit like firebird

Mark Henderson

00:38

@JoelESalas Not far off

Joel E Salas

@MarkHenderson Weee

Andrew

If you have basic math, flow control variables you can do just about anything. won't be pretty, but you can do it.

Matthew Ife

@ewwhite ok work in progress. Dont imagine itll work first time..

ewwhite

@MatthewIfe heh, there's a $1000 bounty on this.

I know that's like £25.

Matthew Ife

lol

HopelessN00b

00:43

@ewwhite That seems unfair. You never put money bounties on fun Windows questions I have a shot at. :p

ewwhite

@HopelessN00b My client called and said that this was worth $1000 to them

Matthew Ife

pastebin.com/fBxwcZJk

HopelessN00b

@ewwhite Well, when you solve it, try to haggle. Ask if it's worth $5000? :)

Matthew Ife

@ewwhite expecting weirdness when exiting, window size changes to not be honoured, other oddities

at this stage, it might not even do the views that it creates

not even tested it

just want to see if this fixes the initial issue.

ewwhite

compiling...

on a test server

Mark Henderson

00:48

@Andrew I'm so confused :( E.G. for the number "3" I want to return 00000011 and for 9 I want 00001001. log(3)/log(2) = 1.58496

Matthew Ife

@ewwhite for 1k, the simplest way to fix this is to ask for the dbc source code, and see if you can just rebuild it to use the system ncurses

which is probably like 15 years later than the shit they are using

ewwhite

@MatthewIfe I'll take that approach tomorrow

today was a "Holiday" in the states

mview.c:(.text+0x4af): undefined reference to `forkpty'

HopelessN00b

@ewwhite It was? I had to show up to work. :(

Andrew

@MarkHenderson do you have a mod operator? and what is your largest possible input?

Matthew Ife

oh sorry

build as

gcc -o mview mview.c -lutil

Mark Henderson

00:50

@Andrew yeah I have mod and I can take any sized input

Andrew

@MarkHenderson do you speak bash/shell?

eh, I'll do some psuedocode

Mark Henderson

@Andrew Yeah I can usually follow that

@Andrew Thanks :(

I failed maths in year 11, year 12 and at university (as you can probably tell)

Andrew

do you have pow?

Matthew Ife

university is for pussies.

Only got GCSEs :(

too poor and stuff to go to real mans school.

Joel E Salas

@MatthewIfe aka people with determination, money and some sense of purpose

none of which I had at 18

ewwhite

00:53

@MatthewIfe It works.

Matthew Ife

heh

BUT is it screwed on exit

ewwhite

It twerks!!

Andrew

how many bits is your smallint?

Matthew Ife

if you resize the term it probably breaks

Mark Henderson

@Andrew 7

No wait, 8

Joel E Salas

00:54

Mark Henderson

Because it includes 128 (which is the max value)

Matthew Ife

I can possibly fix that with a signal handler.

Andrew

@JoelESalas That's Devops!

ewwhite

on exit, it dumps the screen

Andrew

@MarkHenderson you want a string out yeah?

Matthew Ife

00:56

thought it might

was it clearing it prior to me breaking it?

ewwhite

but that's okay... users' sessions are trapped and exiting the software kills their connection

Mark Henderson

@Andrew That would be easiest

But if I can just query a bit of the int, I can do the stringifying part on my own

ewwhite

Now a little more interesting...

Matthew Ife

hehe

lemme see if i can clear up the exit a little..

ewwhite

it's just session stuff... not the worst thing

Matthew Ife

00:57

well,

best thing to do

is wrap that in a shell script

on exit of the program do reset

thatll clear it up

ewwhite

it's already wrapped.

Matthew Ife

window resizing is more tricky

more C for that is required

ewwhite

@MatthewIfe we use a modified Putty, which scales the window and font simultaneously...

Let me try that.

HopelessN00b

So... "eval" would be a decent synonym for test, yeah? (As in eval.compnay.com for our new test forest that doesn't live in the root DNS domain.)

ewwhite

@MatthewIfe It's working well...

screen doesn't actually resize... it scales.

Matthew Ife

01:07

ok, happy?

ewwhite

@MatthewIfe I was going to let customer beat up on it in a test environment, just to make sure it worked for them.

Matthew Ife

ok

ewwhite

The exit could be cleaned up a bit

but the intended targets of this routine are warehouse employees on green-screen terminals

HopelessN00b

@ewwhite Seems odd that your green screen terminal has a red background. They willing to throw money at you to change that?

Matthew Ife

@ewwhite suggest the best way forward for cleanup is to send a reset

its way simpler than trying to work through the maze of code

ewwhite

01:11

@HopelessN00b That's in PuTTY... but this is one of the last clients that still has Relisys Serial Terminals...

Ryan Ries

Is that PoS app using vector graphics or something

Mark Henderson

@Andrew - don't worry, I managed to figure it out

Thanks for the help

Ryan Ries

PoS... Point of Sale, or Piece of Shit? You decide...

HopelessN00b

@RyanRies The distinction only matters when you're using a PoS app that isn't point-of-sale.

Andrew

@MarkHenderson oh, I was wondering if you wanted smart or simple

@MarkHenderson pastebin.com/1QUGjUqv

sorry, busy

should really get myself a maple interpreter

Mark Henderson

01:17

@Andrew Yeah thats pretty much what I ended up with. Collegue came back and was able to help me

I was missing the whole subtraction part

ewwhite

@MatthewIfe customer is happy....

how much do I owe you? @MatthewIfe

Matthew Ife

$1000

:P

Put it on my tab.

ewwhite

I"m not kidding.

Matthew Ife

Split it down the middle if you want.

Need to figure a way out you can send me it. Will speak to the mrs tomorrow. I think shes got paypal.

I so wanted the gdb call to work.

ewwhite

paypal is easiest.... it's how I pay @mdmarra

Matthew Ife

01:21

I'll speak to you tomorrow about money. Wife in bed.

ewwhite

kk.

thank you

Matthew Ife

Still would of been cooler if the gdb call set_escdelay(30) trick would of worked.

ewwhite

alas... that's how it is.

remember, this program hadn't been touched since... oh, 1997...

and the ncurses linked into DBC looks like it was from the RHEL4 days

Matthew Ife

I guess you know that the curses doesn't work on modern systems.

Else they wouldn't statically compile it in.

Likely doesn't even build.

ewwhite

@MatthewIfe you've seen the programming language that this is associated with... it's all pretty old.

Matthew Ife

01:25

Well, they still use sigaction for their signals. Not sure how old that call is. I frequently moan at developers who use signal()

albet some of the terminal setting stuff was ancient.

Andrew

Oh dear. My YouTube suggestions are all over the place.

ewwhite

@MatthewIfe Meanwhile, the server is an Intel E5-2643v2 3.50GHz CPU, 96GB RAM, Fusion-io cards, ZFS-on-Linux...

Matthew Ife

lol

Would actually point out at this point I'm still a sysadmin by trade (and consider that my prime function). Knowing a lot about syscalls just kinda helps me get to the bottom of things / point out why their shit is broke.

So I can polish them a new one.

ewwhite

@MatthewIfe When I worked in trading, we'd dig that deep... Homegrown software, LD_LIBRARY_PATH for all of the kernel-bypass fun, dealing with things blowing up in production... but the past two years, I barely even deal with Linux.

Matthew Ife

Heh, our interview process is rough.

I guess that kind of demonstrates why.

ewwhite

01:32

I think certain aspects of DevOps culture also makes that level of knowledge unnecessary...

Andrew

@MatthewIfe is it a "You must complete every question to pass" interview?

Matthew Ife

Not to that deep a level but we almost expect a rudmentary strace and an account of what is going on.

ewwhite

@MatthewIfe Well, of course... most sysadmins know that.

Matthew Ife

We have a test kind of like the issue that @ewwhite had with the open using O_TRUNC and syncing to disk.

The anomaly being (I thought) that that shouldnt happen. But turns out that you always sync your initial writes after a truncate.

The test being why rewriting over the same parts of a file repetitively doesnt at 2g/sec doesnt appear to produce any i/o load.

HopelessN00b

@MatthewIfe I much prefer the "sysadmins admin, developers... fix their broken code" approach to IT, TBH.

Matthew Ife

01:35

I've worked in hosting. If you cant prove its their code. Its NOT their code.

HopelessN00b

@MatthewIfe The advantage of Windows. "Oh, your code errors out? You should fix that, then."

Matthew Ife

I can see that :)

I've done some windows hosting, your pretty much right there. Windows developers seem to just be happy to take it at face value its their fault.

You lucky git.

ewwhite

Now I have to go debug Perl...

Argument "sks" isn't numeric in numeric gt (>) at /usr/local/lib/perl/Orca/SourceFile.pm line 987, <FD> line 2.
Use of uninitialized value $_[300] in division (/) at (eval 9213) line 3, <FD> line 2.
Use of uninitialized value $_[300] in division (/) at (eval 9213) line 3, <FD> line 2.
Use of uninitialized value $_[300] in division (/) at (eval 9213) line 3, <FD> line 2.
Use of uninitialized value $_[309] in subtraction (-) at (eval 9213) line 3, <FD> line 2.
Use of uninitialized value in subtraction (-) at (eval 9213) line 3, <FD> line 2.

lata!

HopelessN00b

@MatthewIfe Not sure why Linux devs don't assume the same, especially given the near-certainty that it is their fault (or the user's fault).

Andrew

@HopelessN00b because "it works on my laptop" has become an "acceptable" excuse

Matthew Ife

01:45

@ewwhite simple. Remove use strict; and all your problems go away.

HopelessN00b

@Andrew Because Linux admins evidently put up with that shit... I sure wouldn't. That's likely elicit a response like "good for your laptop. Let me know when you get it working on my server" from me.

Matthew Ife

I think the fundamental reason ultimately is it is possible to adequately pull something apart from a system level to work out the problem. On windows its all one mystery black box.

Andrew

@HopelessN00b because most linux devs start out hosting on their laptop, move onto some crappy ubuntu install, and by the time they get to hosting assume "but I got it to work, why can't you"?

especially if they're from a background where they've never had to really deal with shared libraries, hardware oddities, low-level stuff etc.

e.g. they learn in Java or (worse) JS

JS is being taught as a first-year CS language now.

Matthew Ife

Also on Windows you dont tend to get blessed as often with shitty duct-tape code which only just works at runtime due to overly forgiving programming languages like PHP that encourage developers to actually be lazy.

HopelessN00b

@Andrew That's depressing. When you put it that, it sounds like Windows admins are better admins than Linux ones, in the general case. Quite the sad state of affairs.

Matthew Ife

01:50

Its a different ecosystem. I dont think they really can be compared.

Andrew

@HopelessN00b Depends who gets given the power. If the dev gets to run his own servers "because you know what you need", that's where it all goes wrong.

HopelessN00b

@MatthewIfe Didn't say they could. Just saying that putting it like that makes Linux admins sound awful, worse than your average Windows admin, even.

Matthew Ife

@HopelessN00b well, I'd disagree obviously :).

Andrew

@HopelessN00b depends at what level. There's just no price barrier to being a linux admin.

HopelessN00b

@Andrew Not really one to being a Windows admin, either.

Andrew

01:55

@HopelessN00b so define "average windows admin"... the kid running hyper-v in his bedroom?

Matthew Ife

Windows by its nature is pretty turn-key. People seem to go with the idea that if the thing doesn't work dont do that thing. On linux its more the attitude that if a thing doesn't work, force it through till it does.

HopelessN00b

@Andrew That guy sounds a little more skilled than your average Windows admin, TBH.

@MatthewIfe Sounds like a pretty good explanation.

Andrew

@HopelessN00b yeah, maybe not a good example

Matthew Ife

@HopelessN00b almost sounds like your saying windows admins understand how to work the system not how the system works.

HopelessN00b

@MatthewIfe Not convinced that your average Windows admin knows either, TBH.

Matthew Ife

01:59

I bathe in your cynicism @HopelessN00b

Andrew

on windows you can't blame the system, because using windows implies a level of fanboyism -> "the system is perfect, must be my program"?

HopelessN00b

@MatthewIfe Well, sounds like a similar state of affairs for Linux admins too, actually. I imagine they're able to get away with it more readily because command lines scare non-technical people.

ewwhite

@HopelessN00b sometimes...

HopelessN00b

@Andrew No, that's not it at all. Wanna hear someone talk shit about Microsoft, take a Windwos sysadmin out for a beer and ask about his day.

I think Matthew Ife about summed it up when he remarked on the different attitudes between the users of the two platforms.

Matthew Ife

@HopelessN00b Depends on where you work too. If you are lucky enough to be surrounded by awesome talented people, your view on what makes a linux sysadmin has a much higher standard.

Andrew

02:04

Anyone have an ASUS router with a USB HDD plugged in?

Matthew Ife

But, frankly. Browse SF and you'll see theres a very large number of individuals who purport to be a sysadmin yet frankly have no clue what they are doing or what impact it might have.

ewwhite

@MatthewIfe sniffle

It's true... but think about it. Most companies can't retain good engineers.

Joel E Salas

@ewwhite I'm impossible to retain

HopelessN00b

So, @ewwhite ... I figure you may have seen this before, but I havea couple ESXi hosts in a cluster showing as 0 second uptime (but up), and no CPU or memory usage. They're hung/frozen, right?

Matthew Ife

@ewwhite Yeah. I can see that being a problem.

ewwhite

02:06

@HopelessN00b Log onto the shell... restart the management services with: /sbin/services.sh restart

HopelessN00b

@ewwhite Certainly true of every company I've worked for.

ewwhite

@JoelESalas you're hungry.

@MatthewIfe and good engineers tend to move on... move up...

I had a really good engineer running one of my produce clients... He left and is now a high level Linux engineer for NASA.

Joel E Salas

@ewwhite NASA don't pay so good

ewwhite

and that's how it goes...

@JoelESalas but the challenge...

he's was a neckbeard......

HopelessN00b

@JoelESalas Plus, government. Sounds like hell to me, but differnt strokes for differnt folks.

@ewwhite So, how do you manage your ESXi hosts? Do people really leave SSH and ESXi shell turned off, and turn them on through vSphere manually all the time?

ewwhite

02:12

@HopelessN00b I had a few hundred ESXi hosts at the last job...

all had the same root password and SSH enabled.

HopelessN00b

@ewwhite Yeah, I think I'm getting to that point for how often I need to turn on SSH to get on these things.

ewwhite

you can manage an ESXi host without enabling SSH and console...

but dealing with things like this require SSH access

most companies tend to sic the Windows admins on the VMware environment.

Andrew

@ewwhite "engineer" is a privileged word.

most places don't want to employ them, they want an IT monkey

HopelessN00b

Yup... guess I'm not your typical Windows admin, what with my mystical ability to manipulate a CLI.

Andrew

@ewwhite because it has a GUI, right?

ewwhite

02:16

@Andrew yeah, and PowerCLI

HopelessN00b

Yeah, still no joy on the ESXi hosts. Thinking it's reboot time.

ewwhite

@HopelessN00b Hmm... local storage?

HopelessN00b

@ewwhite No, all SAN-connected on this cluster

Andrew

Expect a million how do I set up email server questions... sigh, thanks Ars.

3

ewwhite

@HopelessN00b what is the boot media?

HopelessN00b

02:19

@Andrew Dammit. I'm still waiting for that new protocol promised by SilentCricle and... the other one that Snowden got shutdown.

@ewwhite That's from local storage, yeah.

Andrew

Silent Circle

Silent Circle is a German Eurodisco band that was formed in 1985. It has three members, vocalist Martin Tychsen (Jo Jo Tyson), keyboardist & composer Axel Breitung, and drummer Jürgen Behrens (CC Behrens). History Silent Circle first performed in 1979, but soon parted ways. In the early '80s, the trio reformed and decided to form a more solid group. Hence, that marked the beginning of Silent Circle. Their first single, "Hide Away - Man is Comin'!", gained considerable limelight followed by a number of other successful singles like "Touch in the Night", "Stop the Rain", "Love is just a ...

ewwhite

@HopelessN00b HP servers?

HopelessN00b

@ewwhite Yup. 460c G7 blades.

hipplar

I need a server sanity check if anyone is willing to talk through a problem with me

ewwhite

k

@HopelessN00b just make sure local storage is okay

HopelessN00b

02:23

@ewwhite Actually, looks like at least one of them just now came back after restarting the services a few minutes back. Think I'll give the other one a few more minutes.

hipplar

I've got a site that is running asp.net webforms and the sever was upgraded to a machine with 24 CPUs... The site handles about 200-300 users/per minunte

does anyone else think that's excessive?

Mark Henderson

@hipplar Thats two 6-core processors with hyperthreading

So it entirely depends on what goes on behind the scenes

ewwhite

@hipplar no, that's not a lot

Matthew Ife

5 users a second is pretty quiet

HopelessN00b

@hipplar Yeah, if it's 24 physical CPUs/cores, then, yes, that sounds like a fair bit of overkill. Of course, better to have it and not need it, than need it and not have it. Lots of room for growth.

Mark Henderson

02:25

@hipplar It might be excessive when viewed in relation to the servers workload, but that's pretty much a standard spec server

hipplar

interesting

ok

It's a rackspace server and I'm trying to lookup if it's physical or hyperthreading

Mark Henderson

A workload like that would typically be virtualised these days though

HopelessN00b

@hipplar Probably HT. 2 hexa-core procs with hyperthrading is a very common server config.

hipplar

yeah. it's a legacy codebase and we just keep throwing hardware at it

Mark Henderson

@hipplar Often the cheapest thing to do

hipplar

02:27

true

Mark Henderson

Although if it's single-threaded that won't be any help

hipplar

it's IIS

Mark Henderson

@hipplar Yeah but the app itself its what matters

hipplar

would you rather see have a single big server or load balance it across a few?

Mark Henderson

How many requests per second?

hipplar

02:28

let me get the exact number

Mark Henderson

@hipplar There's no right answer to that

Andrew

@HopelessN00b e.g. a HP DL360 G8

hipplar

@Mark I know. that's why I cam here and didn't ask a question

HopelessN00b

@MarkHenderson Yeah, wish I could my bosses to understand that. Just buy another $8,000 blade instead of trying to get me to spend 200 hours debugging the problem.

Mark Henderson

Depending on how the app works, some don't load balance very well and you have to use sticky sessions. Others you can round robin over 100 nodes without issue

hipplar

02:29

just looking for thoughts

Mark Henderson

@HopelessN00b Yep. Throw another host behind the haproxy instance if you can license it

At $175/hour it doesn't take long for debugging stupid problems to be cost-ineffective

HopelessN00b

@Andrew It's actually because of the popular Xeon processor (Dells, HP blades, other HP models, and more share that CPU config, because they have two of the same CPUs installed).

Andrew

@HopelessN00b indeed

hipplar

1,930 rpm

HopelessN00b

@hipplar Yes, it's overkill, but throwing hardware at the problem isn't usually a bad approach, and I'm not sure I'd want to put something like that on a lower-end server anyway... not to mention the savings might not be worth it.

Mark Henderson

02:31

@hipplar 32 requests per second

hipplar

don't get me wrong... I'm all for buying hardware

Mark Henderson

Depending on what a request actually is, and assuming most requests are only a fraction of a second in processing time, you should be able to scale very high on that

hipplar

well we just went up from 16 CPU because we were spiking

if you guys think it's common I can accept tht

HopelessN00b

@hipplar Yup. Might be overkill, but it's reasonable overkill.

Mark Henderson

@hipplar Spiking a single thread, or spiking overall processor usage?

hipplar

02:34

overall processor usage

HopelessN00b

You know, that said, I have some cheap, old servers I'd trade you for it, if you're worried about the overkill bit. =D

Mark Henderson

@hipplar THat was probably a dual quad core with HT. Assuming you've gone up a generation or two in the upgrade then you will most likely benefit greatly just from the new architecture

Not just the increased core count

hipplar

@MarkHenderson thanks

Andrew

I really hate SNMP as an example of "agentless data collection". If the distro doesn't include it by default, and the default config when it is installed needs to be configured before it gives you what you want - congratulations, it's now an agent!

HopelessN00b

Technically, it's an agent either way... just a question of whether it's pre-configured or not.

Andrew

02:46

cargo-cult IT management.

HopelessN00b

Yeah, I pretty much hate my parents for raising me with ethics and values and other severe disadvantages. Had they not taught me the difference between right and wrong, I could have a career in management. Be a ton easier than working for a living.

Transcript for