|
I am at a similar point and just started similar a kind of research.
We have been running on Fedora for several years now, using ultramonkey,
ldirectord and LVS. It has been very reliable; however - at
least for us - there are two big problems with this combination:
A) Fedora has very short release cycles. It is hard to keep production servers
up-to-date and the point at which a version is not
maintained any more even by the Fedora legacy project seems to be reached too
soon. This is compounded by the fact that we do not
have a hardcore full-time linux-guru sysadmin.
B) Ultramonkey does not come with Fedora, so upgrading the kernel becomes a
major pain because every time a kernel upgrade is
necessary, sources have to be collected from different places, dependencies
have to be resolved, etc.
So I have started some research to see if there is a linux distro that does
come with ultramonkey or something similar, has slower
release cycles and is still free, yet production-reliable.
When I saw your post, somecomments came to my mind. First however, a quick
disclaimer: Since my research is not complete yet, there
might be things that I have overlooked or gotten wrong (particularly packages
in a distro which are there but which I am not aware
of):
1. Because you mentioned ultramonkey: as far as I know, Redhat does not ship
with an ultramonkey RPM (someone correct me if I am
wrong). So you would have to compile from source, make sure everything is
compatible, and do the same thing again each time you
upgrade to a new redhat version or even just a new kernel RPM. Unless you enjoy
compiling stuff from sources and have the luxury of
having enough time and knowledge to do so (or somebody who does it for you) you
should go with something that comes standard with
whatever Linux distro you use, so that the pain and risk during an upgrade is
minimized. Remember - you are talking about production
servers in the end.
2. In case pricing is an issue: CentOS 5 is free and seems to ship with the
entire Redhat clustering suite. I have not used it yet,
but just started to look into it to make sure that nothing is missing in
CentOS. CentOS is Redhat based, so all the config files are
in the same places that Redhat and Fedora have them, there are the same package
managers, services, config-programs, install
screens, etc.
3. I went through Redhat's "Cluster Suite Overview" document and it seems like
they do not offer any file-system level redundancy or
failover. What I mean is essentially something like a Raid-1 or Raid-5 spread
over several servers, which is transparently mountable
over NFS or some other remote file system protocol. So in case one of the
storage servers goes down, the other server(s) still
has/have all the data. I think DRBD would be the closest thing to that, but I
am not sure if Redhat comes with that; at least it is
not mentioned in any Redhat doc I have read so far. I am not sure if CentOS
comes with it...
Redhat does come with their GFS (Global File System), which is essentially a
Logical Volume Manager (LVM) that spreads volumes over
multiple different servers, but it seems like there is no redundancy. Whatever
data is stored on one server is not stored on any
other server. So if that one server goes down, whatever part of the GFS file
system was stored on that server is gone.
4. Gentoo seems to come with a lot of different HA frameworks and packages.
However
I hope this helps. Let me know if you find out more, or if anything I said
above is wrong/incomplete and needs to be corrected.
MARK
> -----Original Message-----
> From: lvs-users-bounces@xxxxxxxxxxxxxxxxxxxxxx
> [mailto:lvs-users-bounces@xxxxxxxxxxxxxxxxxxxxxx] On Behalf
> Of Nick Stephens
> Sent: Thursday, July 05, 2007 10:58 AM
> To: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
> Subject: [lvs-users] LVS software comparisons and opinions
>
> Hi all,
>
> I am quite new to LVS, and learned about it initially as a
> part of Redhat
> Enterprise Servers. Because of that I initially thought that
> redhat was
> the primary resource for clustering and load balancing, but
> thanks to the
> linuxvirtualservers.org site I now see that there are many options.
>
> The one thing that I do not see on the website, presumably
> because it is
> opinion based, is a comparison of the various resources available to
> achieve HA load balancing.
>
> To that end I pose this question:
>
>
> What do you feel is the "best" (in terms of reliability, ease of
> maintenance, etc) option to pursue when attempting to create an LVS?
>
>
>
> As a followup I will pose a pretty simple scenario (mine) in which we
> would use this solution:
>
> 6 redhat linux webservers (rhel5) running apache and resin
> (java). Resin
> handles it's own load balancing internally, so the only real
> issue here is
> apache.
>
> My aspirations are to introduce a solution that allows us to
> have a single
> IP dns entry (as opposed to our current round robin setup),
> and have new
> clients sent to the realserver that is least in use. Another
> critical
> component is monitoring the realserver to ensure that port 80
> is still
> answering, and to stop sending clients there if it isn't. Pretty
> straightforward for this setup, right?
>
> So if you had to set this up and sell it as a reliable HA
> system to your
> boss(es), which software package would you use for this? Piranha,
> Ultramonkey, Keepalived?
>
> Thank you so much, your opinion is important to me. If this
> topic has
> been covered ad nauseum in a place that I have not found,
> please feel free
> to point me in the right direction!
>
> Thanks,
> Nick
>
> _______________________________________________
> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
> Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
> or go to http://lists.graemef.net/mailman/listinfo/lvs-users
>
|
