Thu, 19 May 2011

Trying to learn something new

No cute kitties or kids here, just geek speak, feel free to skip this.

I finally decided I couldn’t keep up with the 6 month new release, 13 month support schedule of Fedora. Way back when I did my server upgrade (about 2 years ago), I planned to install CentOS on the servers, but when I actually started working on them, I just loaded the same Fedora I had on my desktop. I’ve finally corrected that (I think mistake) and have CentOS on both my servers. Here’s a bit of trivia from the process:

By way of background: I have a CentOS PC in my office connected to a DSL modem. After upgrading to 5.6, it no longer talked to the modem — it did talk to the switch in the closet on my floor so I talked Verizon into sending me a new modem. When the new modem didn’t work any better, I booted into the old, 5.5 kernel and that worked fine. Not quite what I wanted, but I didn’t want to spend any more time troubleshooting.

So going into the 5.6 install at home (with a similar DSL modem), I was a little worried about losing my Internet connection.

At about 11:00am I started with trying to install from the CentOS DVD I had downloaded and burned. The SHA1 checksum of the download matched and the disk check part of the install went fine. Unfortunately, the actual install bombed out. No problem, I remembered similar problems with the 5.5 install on my other server. So I dropped back to the 5.6 LiveCD — again the SHA1 checksum and disk check went OK but the install bombed. Not a big deal, only 11:43am by the time I went for the network install CD (checksum and disk check OK).

That was looking fine (thank you Georgetown for the local mirror) until after the stage two kernel was downloaded, partitioning all set and partitions formatted. At the stage when it starts to pull down the packages to install (after pulling down the kernel and the installation packages fine) it bombed. Of course that’s where the message says, “this may take several minutes.” After deciding 30 was more than several, I presumed it was the 5.6 / DSL modem problem and moved on to a network install of CentOS 5.5. About the same time I also remembered one reason for my new Nexus S phone and fired up a local hotspot for my laptop. The Georgetown server was responding fine, even though I kept running into that problem at the same spot in the install (I did try more than once).

When the 5.5 install hung at the same screen, I first thought, “maybe 30 minutes isn’t too many to be several” and I was just being too impatient. Then I fired up Wireshark to check the network traffic on the switch and prove to myself that the switch was OK and I could use the DSL modem with my laptop. All I was seeing was ARP traffic — not a good sign. (Later, I decided my little Dell switch really is a switch and I didn’t see any of the IP traffic because the switch knew not to send it to my laptop’s port.)

It was about 2:00pm, as I was still scratching my head, that I started to give much more serious thought to Ubuntu (I’m running that on my laptop and Robert’s PC; it seems less painful to update and support for the LTS edition is long enough to stay off the frequent upgrade bus). But I really feel more comfortable with SELinux as delivered with RedHat (and hence CentOS) and more familiar with the RedHat way of doing things.

After a break to spend some time with the long suffering family, back to the fight: I thought about trying to run Wireshark on the server in order to see better what was going on with the network — but of course, I didn’t have enough of an OS to actually boot the machine. The answer was this: mount the files from CentOS 5.6 DVD on my internal server and do the network install pointed locally rather than at the nice folks at Georgetown. Two benefits: One it was much faster across the 100Mbs switch than the DSL connection. Two it worked; that was the clincher.

And giving it more thought as the install progressed (and using my laptop through the Nexus S’s hotspot for research), I hit upon the problem: the network stack in the newer kernels were not properly auto-dectecting the 10Mbs DSL connection from the 1Gbs network card. After the install I turned off auto negotiate, set the card to 10Mbs and it connected fine. Actually, before I went down the 5.6 local network install path, I did some testing with an extra NIC card in my internal server. It was running 5.6 and worked so I was pretty sure if I could get 5.6 installed, I could get networking to work for me.

After figuring how how to actually get the OS installed, it was only about 4 hours to install, update and get most of the configuration done (of course, with all the unproductive preamble, it was about 3:30am by then). There were a couple final clean up steps and I ended up wanting to tune logwatch a bit to get a little more detail on sendmail (BTW: Logwatch with Detail set to 5 generated about a 23MB report of one day’s activity. Much better to set Detail just for sendmail to 5 and leave overall Detail at 0 or 1.)

So some keywords that I searched on without luck in case someone else is in a similar boat: centos westell 6100 verizon linux networking 5.6. (I guess you can tell my DSL modem is a Westell 6100 on a Verizon DSL line — default password has been changed, by the way.)

PS: The same force-10Mbs-connection trick worked in my office as well; I’m running the current kernel there now too.

0 comments

comments...

trackback (explanation)

TrackBack ping me at:

https://www.kayakero.net/news/blosxom/file_130585984821508.trackback

Back to News and Updates