http://mirror.ancl.hawaii.edu/linux/centos/
I have excluded obsolete releases, but I am willing to bring back the entire 5.x series if you insist upon it.
Syncing 4 times a day. We have lots of bandwidth, but we would prefer to ramp it up slowly. We are open to the public, but we would prefer to serve mainly Hawaii and the U.S. West region if that would be possible sometime in the future.
mirror.ancl.hawaii.edu is hosted by the Advanced Network Computing Lab at the University of Hawaii, with hardware donated by Netgear and other companies. I am the contact person to administrate the mirror configuration.
Question ======== Have you folks implemented a GeoIP based MirrorManager similar to Fedora Infrastructure? We would like it very much, as we can supply all ASN's and IP subnets of Hawaii networks so all those machines can automatically use the nearest mirror.
Warren Togami wtogami@gmail.com
Am 20.07.11 03:24, schrieb Warren Togami Jr.:
http://mirror.ancl.hawaii.edu/linux/centos/
I have excluded obsolete releases, but I am willing to bring back the entire 5.x series if you insist upon it.
Well. Why not a 6 only mirror :)
I'll add it later today.
Syncing 4 times a day. We have lots of bandwidth, but we would prefer to ramp it up slowly. We are open to the public, but we would prefer to serve mainly Hawaii and the U.S. West region if that would be possible sometime in the future.
That is not possible at the moment.
Question
Have you folks implemented a GeoIP based MirrorManager similar to Fedora Infrastructure? We would like it very much, as we can supply all ASN's and IP subnets of Hawaii networks so all those machines can automatically use the nearest mirror.
We want to/will do that. That is something I wanted to help implement, but I am terribly short on time at the moment, as I am moving between towns during the next ones and begin a new job.
Ralph
On 7/25/2011 11:52 AM, Ralph Angenendt wrote:
Question
Have you folks implemented a GeoIP based MirrorManager similar to Fedora Infrastructure? We would like it very much, as we can supply all ASN's and IP subnets of Hawaii networks so all those machines can automatically use the nearest mirror.
We want to/will do that. That is something I wanted to help implement, but I am terribly short on time at the moment, as I am moving between towns during the next ones and begin a new job.
Currently the vast majority of our mirror traffic appear to be random CentOS clients from around the world. Meanwhile, CentOS clients located in nearby peered networks are randomly hitting worldwide mirrors. This means we are benefited almost not at all by running a CentOS mirror.
Are you able at least to list our mirror at the top for clients asking for a mirrorlist from two particular ASN's?
6360 University of Hawaii 10838 Hawaii Roadrunner
Could you also display our mirror less often for worldwide clients? 50% as often would be ideal.
Warren Togami warren@togami.com
Am 28.07.11 23:21, schrieb Warren Togami Jr.:
Currently the vast majority of our mirror traffic appear to be random CentOS clients from around the world. Meanwhile, CentOS clients located in nearby peered networks are randomly hitting worldwide mirrors. This means we are benefited almost not at all by running a CentOS mirror.
South American people are sent your way (when you are in mirrorlists), as we don't have enough mirrors in .cl, .br, .ar and .cr - these are the lists you're in besides .us
Are you able at least to list our mirror at the top for clients asking for a mirrorlist from two particular ASN's?
Not with our current setup, no.
Could you also display our mirror less often for worldwide clients? 50% as often would be ideal.
You're only getting listed for the two american continents, and that happens on a random basis for everyone.
Regards,
Ralph
On Thu, Jul 28, 2011 at 5:46 PM, Ralph Angenendt ralph.angenendt@gmail.comwrote:
Am 28.07.11 23:21, schrieb Warren Togami Jr.:
Currently the vast majority of our mirror traffic appear to be random CentOS clients from around the world. Meanwhile, CentOS clients located in nearby peered networks are randomly hitting worldwide mirrors. This means we are benefited almost not at all by running a CentOS mirror.
South American people are sent your way (when you are in mirrorlists), as we don't have enough mirrors in .cl, .br, .ar and .cr - these are the lists you're in besides .us
We are in the same rack at mirror.hosef.org, and over the last few years we've gotten traffic from all over. It seems a fair exchange. I believe discussion threads exist in the archives that address this concern via yum for those who want to hard code their local mirror to the local clients they manage.
Are you able at least to list our mirror at the top for clients asking for a mirrorlist from two particular ASN's?
Not with our current setup, no.
Could you also display our mirror less often for worldwide clients? 50% as often would be ideal.
You're only getting listed for the two american continents, and that happens on a random basis for everyone.
If your mirror's traffic patterns remain similar to ours, and you continue to mirror the same distros, you'll find the CentOS traffic is heaviest for its ISO's and DVD's, but is still eclipsed by a few of the other distros we host. For the bandwidth and hardware you have, it should not be a concern. We have served more, with less, for some time now. The 4.x series is still steadily sought.
Regards,
Ralph
Respectfully
--scott
On 7/28/2011 11:46 AM, Ralph Angenendt wrote:
Am 28.07.11 23:21, schrieb Warren Togami Jr.:
Currently the vast majority of our mirror traffic appear to be random CentOS clients from around the world. Meanwhile, CentOS clients located in nearby peered networks are randomly hitting worldwide mirrors. This means we are benefited almost not at all by running a CentOS mirror.
South American people are sent your way (when you are in mirrorlists), as we don't have enough mirrors in .cl, .br, .ar and .cr - these are the lists you're in besides .us
Are you able at least to list our mirror at the top for clients asking for a mirrorlist from two particular ASN's?
Not with our current setup, no.
The University here is OK with providing the worldwide mirror, however they are less than pleased that on-campus CentOS clients are not using the local mirror by default.
Please allow me to help you folks deploy MirrorManager?
Warren
The University here is OK with providing the worldwide mirror, however they are less than pleased that on-campus CentOS clients are not using the local mirror by default.
This isn't a perfect solution but it does work rather nicely ...
If the on-campus CentOS clients use dns resolvers that you control, you could add a zone for mirrorlist.centos.org and direct it at an http server that you run. Then whip up a script that takes (via GET) release, arch, and repo and have it return your mirror.
Thanks, Jef
On Fri, Jul 29, 2011 at 08:25:47PM -0500, Jeff Fisher wrote:
The University here is OK with providing the worldwide mirror, however they are less than pleased that on-campus CentOS clients are not using the local mirror by default.
This isn't a perfect solution but it does work rather nicely ...
If the on-campus CentOS clients use dns resolvers that you control, you could add a zone for mirrorlist.centos.org and direct it at an http server that you run. Then whip up a script that takes (via GET) release, arch, and repo and have it return your mirror.
This works only so long as CentOS uses http instead of https to distribute the mirror list, and so long as yum ignores the https origins of the certificates. Fedora switched to using https a few releases back. I don't recall if yum does https cert checking yet though...
This works only so long as CentOS uses http instead of https to distribute the mirror list, and so long as yum ignores the https origins of the certificates. Fedora switched to using https a few releases back. I don't recall if yum does https cert checking yet though...
Exactly why I said it wasn't perfect but until the state of CentOS mirroring improves -- it's about all we can do.
I'm a long time CentOS user but fairly new to actually following the CentOS project. Who has the power to actually get the ball rolling on using MirrorManager/MirrorBrain or at least move the discussion ahead?
Jeff
On Tue, Aug 02, 2011 at 03:16:02PM -0600, Jeff Fisher wrote:
This works only so long as CentOS uses http instead of https to distribute the mirror list, and so long as yum ignores the https origins of the certificates. Fedora switched to using https a few releases back. I don't recall if yum does https cert checking yet though...
Exactly why I said it wasn't perfect but until the state of CentOS mirroring improves -- it's about all we can do.
I'm a long time CentOS user but fairly new to actually following the CentOS project. Who has the power to actually get the ball rolling on using MirrorManager/MirrorBrain or at least move the discussion ahead?
I have set up an experimental MirrorManager instance for CentOS at http://134.108.44.54/mm.centos/publiclist/ a few months ago and Peter did the same for MirrorBrain at http://centos.mirrorbrain.org/
I am still willing to help to set up MirrorManager if wanted.
Adrian
On 8/2/2011 8:14 PM, Adrian Reber wrote:
On Tue, Aug 02, 2011 at 03:16:02PM -0600, Jeff Fisher wrote:
This works only so long as CentOS uses http instead of https to distribute the mirror list, and so long as yum ignores the https origins of the certificates. Fedora switched to using https a few releases back. I don't recall if yum does https cert checking yet though...
Exactly why I said it wasn't perfect but until the state of CentOS mirroring improves -- it's about all we can do.
I'm a long time CentOS user but fairly new to actually following the CentOS project. Who has the power to actually get the ball rolling on using MirrorManager/MirrorBrain or at least move the discussion ahead?
I have set up an experimental MirrorManager instance for CentOS at http://134.108.44.54/mm.centos/publiclist/ a few months ago and Peter did the same for MirrorBrain at http://centos.mirrorbrain.org/
I am still willing to help to set up MirrorManager if wanted.
Adrian
Matt Domsch wrote some exciting new features including multiple named ASN and netblock support in the upcoming MirrorManager 1.4. It should be in production for Fedora and EPEL this month. After that stabilizes, it should be perfect for CentOS.
Warren
On 8/3/2011 12:05 AM, Warren Togami Jr. wrote:
Matt Domsch wrote some exciting new features including multiple named ASN and netblock support in the upcoming MirrorManager 1.4. It should be in production for Fedora and EPEL this month. After that stabilizes, it should be perfect for CentOS.
Warren
Is the fastestmirror plugin default on CentOS?
If so, that breaks the preferred ordering method of MirroManager, but we could make a simple modification to fastestmirror to solve that issue. We can have MirrorManager put a particular keyword in the topmost comment. If fastestmirror sees that keyword, it knows the mirrorlist is ordered purposefully and just uses it.
So the majority of clients are given random mirrors in their region, and fastestmirror does its connect test.
Other clients are given from MirrorManager a preferred mirror, then fastestmirror knows not to question it. If the preferred mirror is down, MirrorManager knows it is down because it does regular crawling, so it wont be displayed as preferred. Then fastestmirror can do its thing on the random list sans preferred.
Tiny change to fastestmirror can make all the difference.
Warren
Am 29.07.11 00:43, schrieb Warren Togami Jr.:
Please allow me to help you folks deploy MirrorManager?
As already said via IRC: Sure, although MirrorBrain might also be an alternative. I am somewhat out of the loop though until beginning to middle of september. BTW: Matt Domsch and Peter Poeml also read here.
If some guys here want to begin drafting, I am all for it.
http://wiki.centos.org/InfraWiki/Mirrors is what came out of an IRC discussion ages ago.
Cheers,
Ralph
Am 02.08.11 20:36, schrieb Ralph Angenendt:
If some guys here want to begin drafting, I am all for it.
Which obviously also means that we need a way to get the current mirror database into a format which one of those solutions can use. But as it is a rather simple format, I don't see any real issues there.
Ralph
On Fri, Aug 05, 2011 at 02:54:44AM -0500, Ralph Angenendt wrote:
Am 02.08.11 20:36, schrieb Ralph Angenendt:
If some guys here want to begin drafting, I am all for it.
Which obviously also means that we need a way to get the current mirror database into a format which one of those solutions can use. But as it is a rather simple format, I don't see any real issues there.
Adrian Reber has a start with ~15 mirrors, probably hand-populated. A script to convert from your format into MM is a couple hours worth of work I expect, nothing significant.
Am 05.08.11 14:43, schrieb Matt Domsch:
On Fri, Aug 05, 2011 at 02:54:44AM -0500, Ralph Angenendt wrote:
Which obviously also means that we need a way to get the current mirror database into a format which one of those solutions can use. But as it is a rather simple format, I don't see any real issues there.
Adrian Reber has a start with ~15 mirrors, probably hand-populated. A script to convert from your format into MM is a couple hours worth of work I expect, nothing significant.
Yeah, that's what I thought, too (except the auth stuff for mirror owners).
Ralph
On Fri, Aug 05, 2011 at 07:43:55AM -0500, Matt Domsch wrote:
On Fri, Aug 05, 2011 at 02:54:44AM -0500, Ralph Angenendt wrote:
Am 02.08.11 20:36, schrieb Ralph Angenendt:
If some guys here want to begin drafting, I am all for it.
Which obviously also means that we need a way to get the current mirror database into a format which one of those solutions can use. But as it is a rather simple format, I don't see any real issues there.
Adrian Reber has a start with ~15 mirrors, probably hand-populated. A script to convert from your format into MM is a couple hours worth of work I expect, nothing significant.
Yes, it is hand-populated. Is there already a CentOS machine where MM can be installed?
Matt, do you plan to update the epel-6 MM package to the 1.4 branch or would you recommend installing the current 1.3.x package?
Adrian
On Wed, Aug 10, 2011 at 06:42:52AM -0500, Adrian Reber wrote:
On Fri, Aug 05, 2011 at 07:43:55AM -0500, Matt Domsch wrote:
On Fri, Aug 05, 2011 at 02:54:44AM -0500, Ralph Angenendt wrote:
Am 02.08.11 20:36, schrieb Ralph Angenendt:
If some guys here want to begin drafting, I am all for it.
Which obviously also means that we need a way to get the current mirror database into a format which one of those solutions can use. But as it is a rather simple format, I don't see any real issues there.
Adrian Reber has a start with ~15 mirrors, probably hand-populated. A script to convert from your format into MM is a couple hours worth of work I expect, nothing significant.
Yes, it is hand-populated. Is there already a CentOS machine where MM can be installed?
Matt, do you plan to update the epel-6 MM package to the 1.4 branch or would you recommend installing the current 1.3.x package?
Last night I merged pingou's work to port to TG2 into MM mainline (on top of what is currently 1.3.7). I'm hoping to find time this week to go through the ~100 patches that are on the 1.4 branch, and port those conceptual changes to mainline on TG2. Then the testing begins in earnest - it's a good bit of code change. Then we talk about packaging for EPEL etc.
For expediency here, CentOS should start with MM 1.3.x as-is; tell me what works well for you, what other features you uncover you need to really consider MM, and we'll go from there.
Thanks, Matt
On 8/10/2011 3:48 AM, Matt Domsch wrote:
On Wed, Aug 10, 2011 at 06:42:52AM -0500, Adrian Reber wrote:
On Fri, Aug 05, 2011 at 07:43:55AM -0500, Matt Domsch wrote:
On Fri, Aug 05, 2011 at 02:54:44AM -0500, Ralph Angenendt wrote:
Am 02.08.11 20:36, schrieb Ralph Angenendt:
If some guys here want to begin drafting, I am all for it.
Which obviously also means that we need a way to get the current mirror database into a format which one of those solutions can use. But as it is a rather simple format, I don't see any real issues there.
Adrian Reber has a start with ~15 mirrors, probably hand-populated. A script to convert from your format into MM is a couple hours worth of work I expect, nothing significant.
Yes, it is hand-populated. Is there already a CentOS machine where MM can be installed?
Matt, do you plan to update the epel-6 MM package to the 1.4 branch or would you recommend installing the current 1.3.x package?
Last night I merged pingou's work to port to TG2 into MM mainline (on top of what is currently 1.3.7). I'm hoping to find time this week to go through the ~100 patches that are on the 1.4 branch, and port those conceptual changes to mainline on TG2. Then the testing begins in earnest - it's a good bit of code change. Then we talk about packaging for EPEL etc.
For expediency here, CentOS should start with MM 1.3.x as-is; tell me what works well for you, what other features you uncover you need to really consider MM, and we'll go from there.
Thanks, Matt
Although, MM 1.3.x can't do much more useful than CentOS' current mirror system. I personally would wait for 1.4 if deploying it twice is a burden.
Warren
Am 20.07.11 03:24, schrieb Warren Togami Jr.:
http://mirror.ancl.hawaii.edu/linux/centos/
I have excluded obsolete releases, but I am willing to bring back the entire 5.x series if you insist upon it.
I just saw that you have 5.6 - that is more or less the entire 5.6 series as it is present on the mirrors.
If you take a closer look, you'll see that most of the older releases is just links and directory structure.
Ralph
On 7/25/2011 11:54 AM, Ralph Angenendt wrote:
Am 20.07.11 03:24, schrieb Warren Togami Jr.:
http://mirror.ancl.hawaii.edu/linux/centos/
I have excluded obsolete releases, but I am willing to bring back the entire 5.x series if you insist upon it.
I just saw that you have 5.6 - that is more or less the entire 5.6 series as it is present on the mirrors.
If you take a closer look, you'll see that most of the older releases is just links and directory structure.
OK I'll add 5.x.
Warren
Am 26.07.11 00:05, schrieb Warren Togami Jr.:
On 7/25/2011 11:54 AM, Ralph Angenendt wrote:
If you take a closer look, you'll see that most of the older releases is just links and directory structure.
OK I'll add 5.x.
Thanks, that will help people who have - for some reason - set $releasever to a point release in their configs.
Mirror has been added, by the way.
Thank you for your support,
Ralph
-
Adrian Reber -
Jeff Fisher -
Matt Domsch -
R. Scott Belford -
Ralph Angenendt -
Warren Togami Jr.