Andy,

For the isos - totally, people should be using more and more p2p formats.

For the yum targets, that is hard to do. Over the years its been proposed many times ( on the yum-devel list ) that someone should come up with a usable plugin that does p2p / bt of some sort, but when it comes down to implementation, it always falls apart. I think delta-rpms tend to solve that problem to some extent, but not everyone is using that - since in many cases, network is cheaper than i/o.

- KB

Karanbir,

Is there any legitimacy in asking people to use the torrent network for this? Surely with such high levels of bandwidth, more nodes with p2p connections are more suited?

Ricardo,

We tried to get Video going - but the equipment wasent upto it. This is something I want to try and solve for the next few events.

Ofcourse, the really cool option is to just organise a local Dojo where you are!

Man, I would love to attend but I can't.

Are there going to be video records, slides or something for those of us who aren't able to attend?

Thanks!

R u using virtualization? If yes .make sure the one u choose is compatible with all the hardware

Karanbir,
These are the components I'm successfully using:
GIGABYTE MB Sc AM3+ 970A-D3, AMD 970, 4xDDR3
4x DIMM DDR3 8GB 1600MHz CL11 512x8 ADATA
Sapphire VGA ATI Radeon HD 5450 HM, 512MB DDR3 VRAM, 64-bit, 650/800, DVI, HDMI, VGA, PCI-E bulk
CPU AMD FX 8-Core FX-8120 3.1GHz 16MB cache 125W socket AM3+, BOX
2x INTEL PRO/1000 PT Desktop Adapter , PCI Express, (Full i low profile)

Node like this works pretty fine. BTW it is also capable to run ESX 5.0/1.
Regards,
David Hrbáč

Sergio, I'm going AMD here, the intel platform is far too expensive to be worth while.

w.r.t Buying the motherboards, this is a Gigabyte SKT-AM3+ 970A, so wherever I can get a reasonable deal is OK. For the 8 core cpu's i was considering the AMD fX8150, ram from crucial and corsair 430watt psu.

what would you suggest ?

If you do not mind, where are you getting those components? Did you take a look to the X79 boards from Intel? What do you want to install on it?

I seem to have miscalculated.. at £190 a node, i can only get 16GB of ram not 32GB. So maybe getting the 8 core AMD's and reducing node count from 4 to 3, with 32gb on each might be a better, more cost effective option for the same core / ram count.

- KB

Matthew,

The HP MicroServer is an AMD N40L, its a dual core at 1.5 Ghz and has 8 gb of ram. As an object store it should be fine in performance terms. And should also handle logging for the instances without a problem. The base OS images are just under a gig or so, and for now I'm quite keen on keeping them local to the instance.

w.r.t hosting - under the table! I'm not going to buy ATX cases's for these, just pile them up one on top of the other, with a couple of inches of headroom for each 'layer' for circulation.

Adam,

Both of those are US based setup's - and unlikely to match these sort of prices.

Does the HP Micro Server have a good CPU or is that going to be a bottleneck?

How about a RAID of the SSDs being served over channel bonded network links to the compute nodes?

And where are you going to put it - thought you had downsized your local installation?

I've feel looking at some options recently; please note both
www.buildablade.com
and
www.atxblade.com
for reasonably non-proprietary options for rack-mounting generic ATX/ITX gear.

Ohh yeah - makes a lot of sense. I've done it awhile back - removed my facebook, myspace, twitter profile - what a waster of time it was. Whoever wants to contact me can do so through email.

Old stuff can be seriously cheap - I've got a 14 IBM blades + enclosure for under 500 pounds...

I am making some tests with IPv6.

What I learnt so far is that Windows & Linux OS/kernels are ok.

Linux daemons are quite ok, just few very minor issues.

Applications are the bis issue because are (and IMHO will be) the weak ring of the chain.

--- 711,014 unique IPv6's

Would be interesting to split this into PE address and "FF-FE-normal" autoconfig adresses.

This layout still rocks.

So it turns out that searching on a couple of different search engines ( not google ) for a simple set of keywords, not the ones I'd look for when seeking a phone number, but understandable none-the-less, turns up my mobile number!

mystery solved, thanks to Major Chadha who went through the effort of tracking me down and then calling me to confirm!

for many people "search on google" is equivalent to "use a computer connected to the Internet".
I mean it means nothing.

I usually tend to ask people where they got the number from, yes - and in every case the answer has been 'search on google'

Ask them where they got your number from. :)

Why didn't you ask those hundreds of people how they got your number?
I'm sure some of them would tell you. And then maybe you can fix that.

Great can I have a multiple identity for a single host like

host github
hostname github.com
user git
identityfile /root/.ssh/viren , /root/.ssh/id_rsa

Could answer that please

+1 Thanks! Every now and then we need to test some old releases - this'll save me manually checking the repositories or JIRA!

We also must ask what is his/her computer skill or education? Some people has no IT background but trying to mastering computer in a nightpao!

We are trying to deploy our product to Amazon EC2, which is based on CentOS 6.0. Currently, it works smoothly. But we have not tested the reliability and performance yet because we still focus on bugs in our features :)

BTW, I am defeated by the two words' validation.

thank you so much for posting this, it fixed the issue I was having with my setup. now I can have a bunch of keys individually configured for my needs.

We have a couple of server on CentOS deployed in our envoirment.

mysql> SELECT count(*) FROM servers WHERE os LIKE 'centos%';
+----------+
| count(*) |
+----------+
| 1333 |
+----------+
1 row in set (0.01 sec)

Peter,

We are going to try and settle into a two weeks from upstream release plan from now. If for any reason it looks like release is going to slip, then we just reactivate the CR/ repo and get important updates through that way.

You should have seen 5.8 already though ( I apologise for the rather late reply from me, the blog software seems to have a mind of its own when it comes to notifying me of new posts ).

- KB

Mike,

This is the process we are now followig :
- 1 week or so, before a new release we copy the old version tree to vault.centos.org; so its already there on release day for new version

- on the day of release, both the old version and the new version should exist on all mirrors, but the symlink that points to the tree ( ie: /5/ or /6/ as opposed to /5.7/ or /6.2/ ) will flip to the new version. Keep that in mind when you setup rsync's or mirror reposync's.

- About two weeks after release day we will drop the older version from mirror.centos.org / msync.centos.org

Also, if you use rsync - you should be able to limit the max number of files that rsync will delete in one run. That should further reduce the impact from an accidental tree delete ( setting it to 20 or so is a good bet ).

- KB

"I am often surprised by the sort of questions asked in the forums or on irc around open source projects - it just feels as if people are going out of their way to inflict pain and suffering upon themselves by trying to find the most awkward and most complicated way to do things"

Not to sound snarky. But with the deplorable documentation that accompanies SO many open source "projects" it's a wonder anyone can get them to run at all. Take the "documentation" for Bacula, Amanda or even tigervnc. It's a joke. The docs are unreadable(or non-existent). And I must say that these "mom's basement" projects annoy the crap out of those of use who have to put software into production. The time Ive spent trying to get some of these cobbled together software packages to run more than offset the costs of buying commercial software. So... the take home message? If you're not documenting your software effectively (or at all) then you should be dragged from your office and beaten to a pulp.

Do you have any sort of guesstimate of when CentOS 5.8 might be released?

I wouldn't say that people are going out of their way, but they are misguided. When you are trying to solve a problem, you get tunnel vision, and may not notice other solutions if you just focus on one bit.

What you suggest is what I call "What is the question behind the question?" What is the original itch you are trying to scratch? and you are right, often times it has nothing to do with question being asked.

Keep up the good work!

Sean

For those of use who keep a local mirror of CentOS, should we go ahead and freeze our 5.7 tree, or are there going to be updates for 5.7 until 5.8 officially releases for CentOS?

I just want to avoid accidentally wiping out my local 5.7 tree when y'all eventually move it to the archives :-) (happened to my 6.0 tree when 6.1 came out)

Ts, what details would you like to know ? we just run a stripped down instance of Plage-builders for CentOS-4 and 5 which in turn use mock

I am curious about the CentOS-5 buildsystem, could you provide some details?

That thing with the CPU - yes, that is a thing that has to do with I/O. I have desktop that I decided to upgrade in order to use for video rendering. Before I upgraded, I had a dual core Pentium CPU @2Ghz. When I was rendering both cores were maxed out. When I upgraded to a Quad Core Q9650, I was expecting much faster rendering times - obliviously. But guess what - rendering times did not change very much - I got a 10% performance increase at most. Turned out that all 4 cores were running at about 35-40% at most, while the HD was struggling to keep up with the data flow. Now I need an SSD too :)

Bhaskar, thats a good doc to point people at - but its too long, and unless someone is really looking to make the effort, and willing to put in the required time, they wont read it. if they skim it, they wont parse it.

What we need is something like that document, only briefer and perhaps something that is friendlier to people reading English but not using English as their native primary language.

Interesting about the power thing, I guess the increased cpu cycles probably explain this.

When you face some stupid or lazy questions please direct them to this page and ask them read and reread it . It's been a page we have been following for quite sometime..


http://www.catb.org/~esr/faqs/smart-questions.html#intro

Hope this will gives them heads up about their query..I generally ask for steps he made before he/she come to me..as I less approchable? No.. I am not..but I am trying to save some of my invaluable time too...Gentoo forum would be such an example as well as Arch...

Heck! people called it attitude..and I told them yes I do have..

Thanks,
Bhaskar

Have you tried using the laptop with the noop I/O scheduler since you have a SSD? Depending on your usage, your overall I/O should improve even further.

Unfortunately, this means more batter drain for you :P

We got the same issues; overheating and slower access on Laptops with Solid State Drives running Centos.

Scientific Linux and Fedora definitively load and run slower than it should be, on laptops with SSD, but I did not check the temperature of the SSDs.

Running Debian Squeeze on SSD laptops shows significant higher speed and absolutely no heat from SSD..

I thought it may be the kernel versions, but Fedora has newer kernel than Debian, but it is still slower, however I did not check the temperature of the SSD running fedora.

It seem that all RedHat derivatives shows a slowing down of speed and heated SSDs.

Since we were not aware about this issue, this is just a empiric report, but we can do more rigorous quantification and analysis, to draw the right conclusions.


Meanwhile, if anyone has solutions or explanations, it would be great, I would like to know them.

Manu,

I have ext4/discard on - afaik, that should take care of the trim stuff.

Sam,

Do you have some stats / details to share about the power issue ?

Surprisingly (to me and you at least) most SSD eat a lot of power, even in idle mode.
The manufacturers just ignored power consumption and are still starting to wake up from this mistake :(

Ah, I was thinking I'd get an email if you replied to this thread, but I wasn't sure so I checked manually (I would have forgotten all about it otherwise). I guess you have that turned off.

Could the CPU thing be due to TRIM ?

yes, the 4 cores running at higher clock rate is exactly down to most things now being cpu bound rather than i/o ( and in some cases on Network.

I use Vim as well, I use Eclipse as well. But mostly I like Jedit, it works. Only thing missing really is a good git plugin

The 4 cores running more often is possibly because they are spending much *less* time waiting for the I/O to be ready (iowait)? Not a performance guru (as you know :-) but just my guess.

Interesting that you are running jEdit? What were the reasons behind that? Was vim not good enough?

Did anyone already try Planthat? A clone of Pivotal Tracker but free. At least currently ;)
www.planth.at