@sceptics:
In my case the "unlimited" in "unlimited bandwith" is real, My friend updated all 300 pc's under his command to 5.5. He used the boot.iso for every one, meaning that all pc's seperatly downloaded the installation. It caused more then 3TB of bandwith usage! And he even gave me a thx for the high download speed!
-------------------- Roelf Software software.roelf.org www.roelf.org --------------------
Op 7 apr. 2011 om 19:30 heeft florian@gruendler.net het volgende geschreven:
OMG, this is notorious!
Long time I have been listening but now I feel urged to talk and release some frustrated thoughts in public.
- To all the new kids on the block who constantly ask for getting "direct"
access to DVD-ISOs the second day after having their mirror set up
STOP asking! The content delivery network of CentOS must stay hierarchical to be efficient. The master volunteers in the team know to identify which networks/hosts have the necessary reputation, technical know-how and capacity to serve other downstream mirrors in a certain are over rsync which ultimately act as proxies to serve files to on-net/directly peered network hosts. It would almost be recommendable that only very few master servers were selected by the CentOS team to publishes new content on and push it one at a time to such downstream servers or downstream servers are given a time window to pull in order to avoid overload on the masters. It's like DNS root servers, if everybody insisted to only get name resolution done by them, it would get bad for everybody while in fact not a single tick more accurate. In a digital world, a digital copy is as good as the original.
- Backbone bandwidth is one thing and sustainable throughput are clearly
measurable resources as how many Megabytes can be pushed out per time unit
Are you guys aware that the data transfer I/O rate from the storage medium is limited. Plain magnetic PATA or SATA disks @7200 RMP or such slow turning disks in a RAID1 or RAID5 array can only read from the disk at a maximum speed of about 1 Gbit/s (disk to buffer) and it gets MUCH worse for concurrent requests when the tracks are remote for the spindle or the buffer runs constantly full. SCSI/SAS disks @10k or 15k RMP and more redundant RAID arrays with a good RAID controller do somewhat better as they are made to concurrently seek and deliver bytes from the disk to the buffer. SSD disks are a much greater thing, and so it their price. A 99$ "super-dedicated-flatrate-rootserver" machine rented from some provider and having some spare diskspace and bandwidth are NOT suitable to be high up in the hierarchy ever, much less can they take the load of thousands of CentOS machines, initiating a yum update at any non-predictable moment in time. Cheap machines from cheap datacenters tend to have "best effort" bandwidth and the bw is commonly shaped if the hardware isn't maxed out before that. The owners are the first ones to scream if their unimportant website served from the same machine gets slow. Be welcome to donate your bandwidth at your level but don't try to impress us with multi TenGE bandwidth and "Tier1" blabla (the autonomous system to which your IP is connected might have that OVERALL, BUT YOUR SERVERS DON'T) and the silly "unlimited" talk. It's more honest to have a dedicated vintage machine (which 5 years back was a top notch iron) on a port limited to SUSTAINABLE 10 Mbit/s. I don't want to discourage anybody from bearing a share of this CDN project but I urge everybody who may be concerned to refrain from seeing this as a boys game: who has the bigger one, who can do it faster and who is cooler.
- The ones who reported that 5.6 hasn't made it yet to their mirror and
increased the rsync frequency near to one Hertz
Stay calm! The world lived without CentOS 5.6 for a good time and it will not stop to turn if it takes a couple days longer. For the ones who live after the time=money paradigm, they have the money to buy RHEL or Oracle Linux, both had their products on the shelves a couple month ago. Sleep one or two nights over and 5.6 will be bright and wonderful on your mirror as well.
- My message to the people behind the CentOS CDN
You do a great job! Thank you for the patience you have with newbies (we all were one day). Go for some known to be working software like Mirrormanager as soon as you can because you lack the time to reinvent the wheel. And please don't forget a few commercial aspects not so evident to people who associate the Internet with their cable or dsl modem. Peering traffic is free (not settled with cash), transit traffic varies extremely depending on the supplier network. I am sure datacenter connecting networks were more likely to kick in if they could control which ASNs get unrestricted access and which ASN transit paths stay out. Such a GUI controlled tool, would make it easy for every admin to change mirror preferences at their own discretion without justifying in public why they had to close the mirror down (probably because they couldn't afford the 99$ anymore) and wondering why they are not ranked No.1 (because they think everything is like SETI@home).
I feel so much better now that I have told
Good evening, good night, or good morning to everyone out there listening this channel ;-)
Florian
-----Ursprüngliche Nachricht----- Von: centos-mirror-bounces@centos.org [mailto:centos-mirror- bounces@centos.org] Im Auftrag von Roelf Wichertjes Gesendet: Donnerstag, 7. April 2011 16:59 An: Mailing list for CentOS mirrors. Betreff: Re: [CentOS-mirror] 5.6 is coming closer
I've got the bandwith and speed for it, I once pulled all 7 install iso's (at the same time) from a high ranked mirror, it took me 16 min. Then did the same but now used my mirror, it took me 15 min. My bandwith is enough. Ps. I know what a busy server is. I get a bonus from my host if i have a lot of traffic. I forgot the tier of the site and i removed the --delete flag. (my mail software is broken and the webmail i'm using can't handle the e-mail format of the list-admin, so i can't read the mail in which he told me the tier).
I'm ready for it!
Roelf Software software.roelf.org www.roelf.org
Op 6 apr. 2011 om 20:42 heeft "Paul Stewart" pstewart@nexicomgroup.net het volgende geschreven:
I'm curious as to what a busy mirror is...
We are currently delivering about 60GB a day of CentOS files... does that put us at the bottom or near the top? ;)
Paul
-----Original Message----- From: centos-mirror-bounces@centos.org [mailto:centos-mirror-bounces@centos.org] On Behalf Of J.H. Sent: Wednesday, April 06, 2011 1:37 PM To: Mailing list for CentOS mirrors. Subject: Re: [CentOS-mirror] 5.6 is coming closer
On 04/05/2011 01:52 PM, Karanbir Singh wrote:
On 04/05/2011 12:30 PM, Roelf Wichertjes wrote:
Maybe a idea, Why not choose the least busy ones, Say there are 100 mirrors, 5 are busy and 5 almost unused Isn't it a better idea to let the 5 busy and the 5 unbusy pull from
centos.org
And have the other 90 pull from the 5 unbusy? That should even the load better.
On 04/05/2011 10:20 AM, Prof. P. Sriram wrote: > Maybe it's been discussed before, but would it not be worthwhile
to
do a
> DNS based thing for this? We create a temporary rsync source
domain
Thats quite a lot of work, I'm more keen on having ACL's in place
that
only allow some specific mirrors ( maybe the 100 busiest ones ) to
pull
from centos.org; and have everyone else pull from them.
from 'busy' -i meant more like kernel.org / heanet.ie or
mirrorservice.org
- KB
Tiering the mirror distribution is pretty common, and honestly makes things a *LOT* easier for everyone. I agree with the sentiments
already
stated, automation is what makes this all doable. Removing things
like
--delete from your mirrors, is just a PITA. Yes accidental upstream removals will happen, but if the mirror infrastructure is structured well it will propagate out and the fix will propagate out quickly.
The way I've normally seen it is a small number (say 10) mirrors are allowed to pull form the master machines, and servers are then encouraged / forced to pull from those tier 1 mirrors. This means
the
tier 1's can pull more often from the upstream, and everyone else can make better use of the 1 & 10 gbps (and associated big hardware)
links
some of the bigger mirrors have. Personally I think it's worthwhile, and it's not too hard to implement.
Keep in mind that the 'busier' servers (kernel.org at least) are in a better position (hardware / bandwidth) to support a greater number of people pulling from them. I would guess many of those "unused"
mirrors
may not be able to support the deluge you could potentially be
pointing
at them, this isn't universal but it's something to be aware of.
- John 'Warthog9' Hawley
CentOS-mirror mailing list CentOS-mirror@centos.org http://lists.centos.org/mailman/listinfo/centos-mirror _______________________________________________ CentOS-mirror mailing list CentOS-mirror@centos.org http://lists.centos.org/mailman/listinfo/centos-mirror
CentOS-mirror mailing list CentOS-mirror@centos.org http://lists.centos.org/mailman/listinfo/centos-mirror
CentOS-mirror mailing list CentOS-mirror@centos.org http://lists.centos.org/mailman/listinfo/centos-mirror