New server setup checklist

[Simetrical]

Steps 1-3 and possibly 4-5 will need to be done by GED, I'll most likely do most of the rest. I need to research setting up RAID and LVM some more: I've never set up Linux software RAID, and while I did set up LVM for the current server, the way I did it was . . . peculiar.

1. Get all parts and put them together (duh). This is done, except that only one CPU is currently working (not sure if that's intentional).
2. Install Ubuntu Server Edition 8.04 LTS x64 on one disk with no RAID. (Maybe. Or maybe we should set up RAID and LVM at this point. I should try out the Ubuntu installer and see what options it gives me.) It's installed with LVM, but no RAID yet.
3. Make sure sshd is running and set up an account for me. Done; I'm aryeh instead of simetrical.
4. Set up LVM ― possibly including for the root partition. This is best done in single-user mode, and it will likely need to be done by GED on-site. I did manage to do it remotely for the current server, but it's kind of ridiculous. We might want to try starting up sshd in single-user mode with no filesystems mounted and see if I can log in and do it that way, though. Or we could set up LVM on the second disk, copy everything from the first to the second, and reboot to the second, then set up RAID 1. Or maybe we should do this during installation, if possible. Done during installation.
5. Set up software RAID. We might need to do this before, after, or concurrently with the previous step: I'm not sure. Or during installation.
6. Make sure the server is named properly and DNS is set up. After a quick glance at Wikipedia, I think I'll call it thor. (Our old server was named loki, and I named the current one odin. Apparently Thor is Odin's son, and he's supposed to be really strong, so as good a fit as any for our new beefy server.)
7. Now the low-level stuff is done. A whole bunch of stuff can now be done in more or less any order:
   • Set up a cron job to alert people of RAID failures. This happens by default in Ubuntu.
   • Make sure lighttpd , php , XCache , etckeeper , logwatch , some kind of mail server Postfix is Ubuntu default , and a hardware temperature checker doesn't work with K10 are all installed from packages and running.
   • Install Percona's version of MySQL 5.1. I'm not sure whether to use highperf (more custom patches to be careful of), or to avoid it and restrict MySQL to four cores or so. I'm inclined to the latter, I doubt we'll ever need more than four cores for MySQL. Ended up with MySQL 5.0, albeit still with Percona patches, because I found that packaged. Restricted thread_concurrency to 8.
   • Set up backups. Backup scripts have been copied and rsnapshot enabled. We should double-check that these actually work.
   • Copy over custom configuration. Everything seems to be copied.
     
   • Copy over /var/www, and load a database backup for testing from mysqldump. We may want to experiment with parallel loading scripts to speed this up: I don't know how long it will take. If it takes too long, we might need to reconsider. rsync works fine for the MySQL data; everything is copied (and will be re-copied before the site goes live).
   • Set things up to boot from both mirrors.
   • Install and configure Sphinx.
   • Set up FTP accounts and inform everyone of their new login info.
   • GED wants to try to get the temperature sensor working.
8. Test RAID: pull out a drive while the server is running and see what happens; try rebooting in that state.
9. Once everything seems to work, make sure our hosting is in order, switch DNS to their DNS servers, cut back the TTL to 15 minutes or so, move thor on-site, and make sure everything works.
10. Make sure mail works. This is tricky to do before this point, since mail @twcenter.net still goes to odin, so we have to do it later in the process.
11. Put up an informative "we're moving the site" static HTML page on thor, then on odin. Take a final database backup from odin, move it to thor (rsync should make this relatively quick), and load it. Make sure userids are correct (they're different on odin and thor in some cases). Turn on the board on thor for some quick final checks by admins, then open it up to general posting. Hopefully this step will take under an hour.
12. Post-move steps: Set lighttpd on thor to redirect everything to www.twcenter.net again (this had to be disabled for testing). Post notices prominently saying that we allow uploads up to 1G now.

The list is subject to change as we think of more things that need to be done.

[GrnEyedDvl]

RAID might be best done on initial setup, but we can go either way on that. I generally do it that way but it doesnt really matter to me. Seems to take less time.

Once the machine is up and running I will install sshd and make sure everything goes through my network like it should so you can have access.

The name thor will work, nothing on my network is using that name. Good thing you didnt pick zeus or ares though

For user accounts we also have to have one for the host in case we really need them to get into it, and they will probably have to change some configurations to set it up in their network when we move it. I will probably just create 1 user account at install, and give you that user name and password so you can create whatever else you need.

To run on my network its going to have to have an internal IP of 10.100.128.xxx so my router and DNS server will forward traffic to it.

I think restricting MySQL to 4 cores is probably a good idea, make sure we have plenty for the cgi processes.


Other software I want:
That hardware package that checks temps, thats pretty useful. Its running on my Ubuntu server just fine. I also want to run the crap out of it for several hours while I am sitting here and make sure we have no issues. I dont expect any, but better to catch it early.

[Simetrical]

Originally Posted by GrnEyedDvl

RAID might be best done on initial setup, but we can go either way on that. I generally do it that way but it doesnt really matter to me. Seems to take less time.

I'll make a note to try installing an Ubuntu server here in a VM to see how the installation process goes. We'd want both LVM and RAID1.

Originally Posted by GrnEyedDvl

The name thor will work, nothing on my network is using that name. Good thing you didnt pick zeus or ares though

I don't think it really matters in terms of networking what name is used, for Linux. Computer name seems to be used mostly in the names of some files, in log events, and on the terminal prompt (e.g., "[simetrical@odin ~]$", "[simetrical@thor ~]$" so you can tell where you are at a glance). My impression is that the computer name in Windows is mainly relevant for networking.

Originally Posted by GrnEyedDvl

For user accounts we also have to have one for the host in case we really need them to get into it, and they will probably have to change some configurations to set it up in their network when we move it. I will probably just create 1 user account at install, and give you that user name and password so you can create whatever else you need.

That's what I was thinking. We'll want the uids to match the ones on the current server, so that file ownership remains correct when everyone's home directories and things are copied over.

Originally Posted by GrnEyedDvl

To run on my network its going to have to have an internal IP of 10.100.128.xxx so my router and DNS server will forward traffic to it.

Wait, it won't have a publicly-routable IP address? It's behind NAT or something?

Originally Posted by GrnEyedDvl

I think restricting MySQL to 4 cores is probably a good idea, make sure we have plenty for the cgi processes.

That doesn't make a difference. It's not like the cores are divvied up among processes in advance. Restricting MySQL to four cores means telling it not to use more than four threads for some things, to avoid too much lock contention, since it doesn't scale well to a lot of CPUs. The operating system will still schedule all threads on the first available CPU as usual (taking into account all the other things it takes into account to avoid arcane stuff like TLB flushes).

Originally Posted by GrnEyedDvl

Other software I want:
That hardware package that checks temps, thats pretty useful. Its running on my Ubuntu server just fine. I also want to run the crap out of it for several hours while I am sitting here and make sure we have no issues. I dont expect any, but better to catch it early.

Shouldn't be a problem.

[GrnEyedDvl]

Originally Posted by Simetrical

I don't think it really matters in terms of networking what name is used, for Linux. Computer name seems to be used mostly in the names of some files, in log events, and on the terminal prompt (e.g., "[simetrical@odin ~]$", "[simetrical@thor ~]$" so you can tell where you are at a glance). My impression is that the computer name in Windows is mainly relevant for networking.

Well sorta. Any machine that is networked has to have a unique ID, the TCP and DNS protocols are the same no matter what the platform is. For internal name resolution, all computers on any network must have a unique name, whether the domain controller or DNS server is Windows or Linux. If I placed a machine on my network, or any network, that had the same name as another machine, then the DNS records would get hosed. I will give it a static IP address behind my public IP address, and all traffic will be routed there.

The current machines (Odin) setup is:

eth0 Link encap:Ethernet HWaddr 00:18:8B:FD:93:9E
inet addr:74.53.195.226 Bcast:74.53.195.239 Mask:255.255.255.240
inet6 addr: fe80::218:8bff:fefd:939e/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:497294554 errors:0 dropped:0 overruns:0 frame:0
TX packets:759516572 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:85011744152 (79.1 GiB) TX bytes:1008340100908 (939.0 GiB)
Interrupt:169

That 74.53.195.226 is the public IP of TWC, because the machine is exposed directly to the internet and TWC "owns" that IP address. On my Ubuntu machine the file looks like this:

eth0 Link encap:Ethernet HWaddr 00:0C:F1:B7:A9:18
inet addr:10.100.128.7 Bcast:10.100.128.255 Mask:255.255.255.0
inet6 addr: fe80::20c:f1ff:feb7:a918/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:524623 errors:0 dropped:0 overruns:0 frame:0
TX packets:277534 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:87424414 (83.3 MB) TX bytes:46147636 (44.0 MB)

That 10.100.128.7 is the internal IP on my network, not my public IP.

That's what I was thinking. We'll want the uids to match the ones on the current server, so that file ownership remains correct when everyone's home directories and things are copied over.

When I get to that point is there anything specific you want the first account created to be named?

Wait, it won't have a publicly-routable IP address? It's behind NAT or something?

Its behind a firewall and also behind Windows IPSec. You will navigate to it by typing my public IP address into whatever you are using to access the shell. Because its coming in on a specific port number the traffic will get routed to thor.

[Simetrical]

Originally Posted by GrnEyedDvl

Well sorta. Any machine that is networked has to have a unique ID

Yeah, that's its IP address. Nothing else is used in low-level networking. Domain names are used, but that's already at the application layer. TCP and IP sure don't use anything but IP addresses. Only higher-level protocols use other designations, and I'm not aware of any protocol commonly used in Unix that requires unique hostnames, or even uses them at all. I do everything by domain name (which translates to IP address).

Originally Posted by GrnEyedDvl

the TCP and DNS protocols are the same no matter what the platform is. For internal name resolution, all computers on any network must have a unique name, whether the domain controller or DNS server is Windows or Linux. If I placed a machine on my network, or any network, that had the same name as another machine, then the DNS records would get hosed.

They can't have the same DNS name without causing havoc, but they can certainly have the same internal name. It might mess up Windows protocols like SMB, but I don't think it makes any difference for Linux.

The internal name in Linux is just an arbitrary designation, as far as I can tell. I don't think it's used for any network operations at all. It will sometimes coincide with a domain name that resolves to the computer in question, but that's optional: on my home machine, the command "hostname" prints out simply "aryeh-desktop". On odin it outputs "odin.twcenter.net". On another server I have access to it outputs "ubuntu". I'm quite certain nothing bad would happen if you hooked two of them together with the same hostname.

Originally Posted by GrnEyedDvl

The current machines (Odin) setup is:


That 74.53.195.226 is the public IP of TWC, because the machine is exposed directly to the internet and TWC "owns" that IP address. On my Ubuntu machine the file looks like this:



That 10.100.128.7 is the internal IP on my network, not my public IP.

Yes, I know, but those aren't related to the names.

Originally Posted by GrnEyedDvl

When I get to that point is there anything specific you want the first account created to be named?

Doesn't matter, I can always change it later.

Originally Posted by GrnEyedDvl

Its behind a firewall and also behind Windows IPSec. You will navigate to it by typing my public IP address into whatever you are using to access the shell. Because its coming in on a specific port number the traffic will get routed to thor.

Which ports are available? Normally I'd open SSH on port 22, SMTP on port 25, and HTTP on port 80. If those don't work, we can probably pick any we like, although of course we should switch them back when we set up the server for real. I'm not sure offhand if you can receive incoming mail except on port 25, actually . . . if necessary we can disable that to begin with and make sure it's working once the server is moved to the hosting facility.

[GrnEyedDvl]

The internal name in Linux is just an arbitrary designation, as far as I can tell. I don't think it's used for any network operations at all. It will sometimes coincide with a domain name that resolves to the computer in question, but that's optional: on my home machine, the command "hostname" prints out simply "aryeh-desktop". On odin it outputs "odin.twcenter.net". On another server I have access to it outputs "ubuntu". I'm quite certain nothing bad would happen if you hooked two of them together with the same hostname.

We are getting out of the realm of the TWC server here, but say you wanted to map a shared path, using either Linux of Windows.

Linux:

smbmount //servername/share /mnt/srvr -o username=________ password=_________

Windows:

\\servername\share

Windows via script or Group Policy Object:

NET USE I: \\servername\folder /persistent:no.

Then it prompts for user name and password. That servername is the name of the machine, or the virtual name if you are using some kind of virtual host. servername has to be registered in DNS for the path to work, whether you are using Linux or Windows.

When a client recieves an IP address via DHCP it takes the computer name and registers that in DNS as its FQDN (Fully Qualified Domain Name) whether its just a simple name like odin or a name like odin.twcenter.net.

If you assign a static IP, you either have to create a manual DNS entry or set the option to register the address with DNS automatically. Either way the machine name, or virtual DNS name, has to be unique or you will get errors in your DNS logs. If you had two machines named odin, and typed a simple command like ping odin, it would fail. However you can have two machines with the same name if one is joined to the domain and one is not, or they belong to two different domains. For example if you had odin.twcenter.net and odin.twcenter.com that would be fine, though there would be some special routing rules you would have to set up.

Which ports are available? Normally I'd open SSH on port 22, SMTP on port 25, and HTTP on port 80. If those don't work, we can probably pick any we like, although of course we should switch them back when we set up the server for real. I'm not sure offhand if you can receive incoming mail except on port 25, actually . . . if necessary we can disable that to begin with and make sure it's working once the server is moved to the hosting facility.

You will have all the standard ports, through the magic of routing tables and supernetting.

[Simetrical]

Originally Posted by GrnEyedDvl

We are getting out of the realm of the TWC server here, but say you wanted to map a shared path, using either Linux of Windows.

Then you're using SMB. As I said, SMB may need unique names, but no protocol normally used on Linux alone does. On Linux, unless you're interoperating with Windows, you'd use NFS, not SMB, to mount remote filesystems. And for NFS you'd just use a command like

mount twcenter.net:/var/local/export /media/twc

which would mount /var/local/export from TWC as /media/twc locally, assuming that we were exporting that directory (we're not, of course, it doesn't even exist). No mention here of anything but domain name, and the domain name is just a convenient way of writing the IP address: the command

mount 74.53.195.226:/var/local/export /media/twc

would work identically to the above (as long as TWC's address doesn't change, as it shortly will).

Originally Posted by GrnEyedDvl

When a client recieves an IP address via DHCP it takes the computer name and registers that in DNS as its FQDN (Fully Qualified Domain Name) whether its just a simple name like odin or a name like odin.twcenter.net.

If you assign a static IP, you either have to create a manual DNS entry or set the option to register the address with DNS automatically. Either way the machine name, or virtual DNS name, has to be unique or you will get errors in your DNS logs.

But the hostname on Linux boxes doesn't have to have anything to do with the DNS name. On my desktop, the hostname is aryeh-desktop, which is unrelated to my reverse DNS of cpe-72-229-x-x.nyc.res.rr.com. Likewise another server I administer has a hostname of "ubuntu", although its DNS name is tech-artists.org and nothing prefixed with "ubuntu." will resolve to it. It's conventional to set up hostnames that are valid DNS addresses resolving to the current host, but it's not necessary.

Originally Posted by GrnEyedDvl

If you had two machines named odin, and typed a simple command like ping odin, it would fail.

Where "named odin" means "with DNS name odin", yes, but not "with hostname odin", which is different (at least on Linux). Actually it wouldn't fail, you can have multiple machines under the same DNS and they're served round-robin. You'd just get one of them or the other arbitrarily. Trying pinging google.com a few times and you'll see that you get different IP addresses (at least I do).

Originally Posted by GrnEyedDvl

You will have all the standard ports, through the magic of routing tables and supernetting.

Then your other servers can't be running services on the same ports. Otherwise how will it know where to send packets addressed to port 22 (or whatever) with your IP address? The domain name isn't necessarily provided in the request.

[GrnEyedDvl]

Originally Posted by Simetrical

But the hostname on Linux boxes doesn't have to have anything to do with the DNS name. On my desktop, the hostname is aryeh-desktop, which is unrelated to my reverse DNS of cpe-72-229-x-x.nyc.res.rr.com. Likewise another server I administer has a hostname of "ubuntu", although its DNS name is tech-artists.org and nothing prefixed with "ubuntu." will resolve to it. It's conventional to set up hostnames that are valid DNS addresses resolving to the current host, but it's not necessary.

You are confusing internet DNS with DNS routing inside a network. That cpe-72-229-x-x.nyc.res.rr.com is the name of the last router that RoadRunner owns before it hits your cable modem. That router probably has a DNS counterpart on the same 72.29.xxx.xxx subnet named something like ns1.nys.res.rr.com or some variation of that.

Incidentally your cable modem also has two IP addresses, your public 72.229.xxx.xxx and http://192.168.100.1/ which is the default broadcast IP for home networks. Theres usually a config page located there.

If you look at the routing tables or DHCP tables within your router (assuming your router is also your DHCP server) then you will see your machine name next to its IP address. This is why you can ping by machine name.

When you name your machine, in Linux or in Windows, that name becomes the default DNS (internal) name. Like this:




Ares is obviously my server, goliath is my sons machines name (he picked it) and pandora is my Ubuntu machine. Note that pandora is not joined to the domain, so it just says pandora. The ones you do not see are zeus (my desktop), athena (wifes machine), apollo (backup server), or either one of my daughters machines. They are not turned on or they would show in the list too, and my backup server is also its own DNS server (10.100.1.xxx) and then there is a bridge between the two networks. I am connected via remote with my laptop so you dont see that connection either.

The Linux machine broacasts its name for DNS purposes, though you can manually change it through suffixes for any operating system. For external/internet DNS resolution there has to be a manual table entry created on a public DNS server, and you can call that anything you please.

Where "named odin" means "with DNS name odin", yes, but not "with hostname odin", which is different (at least on Linux). Actually it wouldn't fail, you can have multiple machines under the same DNS and they're served round-robin. You'd just get one of them or the other arbitrarily. Trying pinging google.com a few times and you'll see that you get different IP addresses (at least I do).

Thats because you are getting routed to different machines based on traffic, thats in internal Google thing, but you can do that on any real server as well as run two websites off the same IP address. Thats all done with host headers and DNS Balancing.

Then your other servers can't be running services on the same ports. Otherwise how will it know where to send packets addressed to port 22 (or whatever) with your IP address? The domain name isn't necessarily provided in the request.

The only real problem with multicasting and dynamic NAT are ports 21 and 25. My email server is running port 587 which is the "new" standard dating back a few years. Most people are still stuck on 25 for some reason though.

If we need port 21, I will move mine to another port for a while. I am the only one that uses it anyways.

[Simetrical]

Originally Posted by GrnEyedDvl

You are confusing internet DNS with DNS routing inside a network. That cpe-72-229-x-x.nyc.res.rr.com is the name of the last router that RoadRunner owns before it hits your cable modem.

No, that's a separate machine:

$ traceroute google.com
traceroute to google.com (72.14.205.100), 30 hops max, 40 byte packets
 1  10.36.0.1 (10.36.0.1)  11.225 ms * *
 2  gig-4-0-nycmnye-rtr1.nyc.rr.com (24.29.98.37)  10.905 ms  10.955 ms  10.946 ms
 3  pos-2-0-nycmnyc-rtr1.nyc.rr.com (24.29.104.162)  10.961 ms * *
...

10.36.0.1 is our home router, I assume (thus the private IP address). 24.29.98.37 is the first RoadRunner router. The reverse DNS I gave is for my own IP address:

$ ifconfig | grep 'inet addr'
          inet addr:72.229.28.14  Bcast:72.229.29.255  Mask:255.255.254.0
          inet addr:127.0.0.1  Mask:255.0.0.0
$ dig -x 72.229.28.14 +short
cpe-72-229-28-14.nyc.res.rr.com.

Originally Posted by GrnEyedDvl

If you look at the routing tables or DHCP tables within your router (assuming your router is also your DHCP server) then you will see your machine name next to its IP address. This is why you can ping by machine name.

There are no other machines visible on my network, so I can't test this. As ifconfig shows, I don't have a private IP address (other than the loopback). The address I'm giving is my public Internet address.

Originally Posted by GrnEyedDvl

When you name your machine, in Linux or in Windows, that name becomes the default DNS (internal) name. Like this:




Ares is obviously my server, goliath is my sons machines name (he picked it) and pandora is my Ubuntu machine. Note that pandora is not joined to the domain, so it just says pandora. The ones you do not see are zeus (my desktop), athena (wifes machine), apollo (backup server), or either one of my daughters machines. They are not turned on or they would show in the list too, and my backup server is also its own DNS server (10.100.1.xxx) and then there is a bridge between the two networks. I am connected via remote with my laptop so you dont see that connection either.

Well, it's not relevant in my case since I don't have a local network. There is one in my house, but I'm not connected to it, since NAT interferes with some of the programs I use (like IRC) and is generally a nuisance. All the computers that are connected to it are Windows machines, so I don't have any way to check what you're saying in an all-Linux setup.

Getting the latest DHCP RFC, I see that "DHCP also does not address registration of newly configured clients with the Domain Name System (DNS)". What protocol are the computers using to broadcast their hostnames?

Originally Posted by GrnEyedDvl

Thats because you are getting routed to different machines based on traffic, thats in internal Google thing, but you can do that on any real server as well as run two websites off the same IP address. Thats all done with host headers and DNS Balancing.

It's just because the DNS record gives multiple IP addresses:

$ dig google.com +nocmd +nocomments +noidentify +nostats

; <<>> DiG 9.5.0-P2 <<>> google.com +nocmd +nocomments +noidentify +nostats
;; global options:  printcmd
;google.com.            IN    A
google.com.        27    IN    A    74.125.45.100
google.com.        27    IN    A    209.85.171.100
google.com.        27    IN    A    72.14.205.100
google.com.        78310    IN    NS    ns4.google.com.
google.com.        78310    IN    NS    ns3.google.com.
google.com.        78310    IN    NS    ns2.google.com.
google.com.        78310    IN    NS    ns1.google.com.
ns2.google.com.        163952    IN    A    216.239.34.10
ns1.google.com.        342159    IN    A    216.239.32.10

Note that there are three A records returned. When I resolve google.com, it will resolve to one of the three at random. This is a feature of DNS and will always happen when the same domain name resolves to multiple hosts. So it's not an internal Google thing.

Originally Posted by GrnEyedDvl

The only real problem with multicasting and dynamic NAT are ports 21 and 25. My email server is running port 587 which is the "new" standard dating back a few years. Most people are still stuck on 25 for some reason though.

If we need port 21, I will move mine to another port for a while. I am the only one that uses it anyways.

Should be no problems, then.

[GrnEyedDvl]

10.36.0.1 is our home router, I assume (thus the private IP address). 24.29.98.37 is the first RoadRunner router. The reverse DNS I gave is for my own IP address:

Yes but its giving you either the name of a Road Runner router as there is no entry in the DNS table maintained by Road Runner for your machine, or possibly the logical name of your cable modem. Its not going further than your modem and picking up your machine.

There are no other machines visible on my network, so I can't test this. As ifconfig shows, I don't have a private IP address (other than the loopback). The address I'm giving is my public Internet address.

Well, it's not relevant in my case since I don't have a local network. There is one in my house, but I'm not connected to it, since NAT interferes with some of the programs I use (like IRC) and is generally a nuisance. All the computers that are connected to it are Windows machines, so I don't have any way to check what you're saying in an all-Linux setup.

Why would NAT interfere with that?? Probably a configuration error someplace. You have your machine in the DMZ then, or not connected to the other network at all?

Getting the latest DHCP RFC, I see that "DHCP also does not address registration of newly configured clients with the Domain Name System (DNS)". What protocol are the computers using to broadcast their hostnames?

Notice the date on that report you linked to, March 1997. That can hardly be the latest as IPv6 was still a distant dream at the time.

But even back then the DHCP protocols required a unique identifier:

DHCP defines a new 'client identifier' option that is used to pass an
explicit client identifier to a DHCP server. This change eliminates
the overloading of the 'chaddr' field in BOOTP messages, where
'chaddr' is used both as a hardware address for transmission of BOOTP
reply messages and as a client identifier. The 'client identifier'
is an opaque key, not to be interpreted by the server; for example,
the 'client identifier' may contain a hardware address, identical to
the contents of the 'chaddr' field, or it may contain another type of
identifier, such as a DNS name. The 'client identifier' chosen by a
DHCP client MUST be unique to that client within the subnet to which
the client is attached. If the client uses a 'client identifier' in
one message, it MUST use that same identifier in all subsequent
messages, to ensure that all servers correctly identify the client.

Later versions of DHCP automatically take the computer name, translate that into the host name and use it as a DNS tag.


It can also be done through NetBIOS, which is pretty much tied to DHCP and DNS both as it runs over the TCP protocol. Windows has its own version of NetBIOS called WINS which allows you to append names differently than DNS does. I assume that similar to how the Linux host-name operates.

[Simetrical]

Originally Posted by GrnEyedDvl

Yes but its giving you either the name of a Road Runner router as there is no entry in the DNS table maintained by Road Runner for your machine, or possibly the logical name of your cable modem. Its not going further than your modem and picking up your machine.

The IP address is mine, not my router's. I'm not behind NAT, my computer has its own publicly-routable IP address. We pay Time Warner for two addresses: one goes to me, and one goes to a second router that services the other three computers in the house via NAT. I'm guessing the router for our house has no publicly-routable IP address at all (why would it?).

Originally Posted by GrnEyedDvl

Why would NAT interfere with that?? Probably a configuration error someplace.

The problem is the ports that need to be forwarded. IRC existed before NAT became popular, and it uses a lot of different ports for different services. You can connect to networks on a whole bunch of different ports, and there are various ancillary protocols like identd and DCC that use their own sets of ports. It doesn't use a fixed outgoing port with a single port that's transparently selected to receive responses, like most protocols do. I believe FTP is similar, but it's considerably simpler and probably more common, so routers tend to have special rules set up to get it to work.

Anyway, IRC is complicated enough that routers might not bother to support it, or might support it incorrectly or incompletely. Practically speaking, the only easy way to get it to work reliably is to forward the whole slew of ports to one machine. When we set up our network, my sister also used IRC, so we couldn't easily be behind the same NAT. So I got split off to my own IP address. I like that anyway because it's more straightforward if I want to run servers for whatever reason. For instance, I run sshd on my desktop (which I also leave on all the time) so I can access my computer from anywhere if desired. Since I'm not behind NAT, I don't have to worry about conflicting with a server set up by my brother, or about having to forward the port to me at all.

Originally Posted by GrnEyedDvl

You have your machine in the DMZ then, or not connected to the other network at all?

The cable modem is connected to a switch. The switch then has two other connections: one goes straight to my computer, and one goes to a router. The router then has three lines out, connected to my brother's computer, my parents' computer, and a wireless modem. These three share an IP address.

Originally Posted by GrnEyedDvl

Notice the date on that report you linked to, March 1997. That can hardly be the latest as IPv6 was still a distant dream at the time.

Work on IPv6 started in the early 90s. RFC 1883 is the first RFC on the topic. Of course, IPv6 is still a distant dream, so I don't disagree on that point.

Anyway, the original idea was that IPv6 would avoid the need for DHCP, by having the host-specific part of the address (lower 64 bits) be globally unique, for instance by including a MAC address. Then the host could just tell the network on connection what its half of the IP address would be, and the network would provide its half (upper 64 bits), with no negotiation needed. Apparently a stripped-down form of DHCP ended up being needed in practice for one or two obnoxious remaining details, like figuring out the IP address of the local nameserver, but certainly not the same form used in IPv4.

Originally Posted by GrnEyedDvl

But even back then the DHCP protocols required a unique identifier:

That's not the same thing. For instance, my /etc/dhcp3/dhclient.conf has this commented-out line:

#send dhcp-client-identifier 1:0:a0:24:ab:fb:9c;

That's sure not being used in any DNS names. I'm not sure DHCP is to the point, anyway, if we're talking about servers, because DHCP is practically never used for servers.

I'll ask some of the real Linux gurus I know (i.e., one of the people who runs Wikipedia) what happens if you hook together a bunch of Linux machines with the same hostname set.

[Garbarsardar]

Please remember that it would be good to notify Ian as early as possible about the moving date so he can terminate our relation with the current host.

[Augustus Lucifer]

Originally Posted by Garbarsardar

Please remember that it would be good to notify Ian as early as possible about the moving date so he can terminate our relation with the current host.

:hmmm: Like such?



Host-------Ian

[Simetrical]

Originally Posted by Garbarsardar

Please remember that it would be good to notify Ian as early as possible about the moving date so he can terminate our relation with the current host.

Of course. We should be able to keep the overlap down to a couple of days, since initial setup will be done at GED's place. We won't be able to figure out timeframes until we've at least gotten the server running.

[GrnEyedDvl]

I got the coolers yesterday, I will try and get it together and booted today.

[Simetrical]

Okay, great. Make sure you get LVM set up as well as RAID, if possible.

[GrnEyedDvl]

I seem to be having a BIOS issue, I think the board we got had been setting on a shelf for a while and doesnt have an updated BIOS for the 82xx Opterons. The problem with that is I cant flash it, I dont have an Opteron 8xxx-82xx processor. Its throwing me an FF error code which either means bad CPU or power supply issues. As everything is new, and I seriously doubt I got 4 bad CPUs, I am guessing it just doesnt recognize the processors. I spent about an hour on the phone with Tyan today and I am probably going to end up getting an RMA on it as they will not ship a BIOS chip to me directly as I am not a Tyan authorized dealer. I will be on the phone again with them tomorrow.

[GrnEyedDvl]

Sorry I havent been around much the last few days.

The board has been RMA'd to Tyan (Monday), which may take as many as 14 days. We cant really wait that long. We have to give a 30 day notice to our old host, and I dont want to gove them notice until the new machine is up and stable. That puts us up against the clock for the release of Empire. A 30 day notice for the Planet puts us in mid February, if I got the machine up and running today. Waiting a further 14 days puts us right at the end of February and I am not comfortable with that cushion.

Sooo... Today I ordered a 1U barebones system that will accept the 4 processors and RAM we already have. This system was not out when I ordered the original parts, damn them. It came out last week. Its a case, board, power supply, and rails with 4 hot swap drive bays at $1100. I am the only person that has ordered this from NewEgg, possibly the first person to order it ever. Since I have never had a problem with a Tyan board until building this machine, I am not too worried about more problems.

Because of this thread, I also ordered another 8 gigs of RAM, so we will have 16 total. That was another $230, but I want to eliminate as many problems as we can here.

All told I spent $1400 that I was not expecting to spend, and had it shipped next day air for $100. That should put it here Friday at the latest, possibly tomorrow if they get it out on time.

I will end up with a case, power supply, and a motherboard that I really have no use for but cannot return. Such is life I guess. I will figure out what to do with the extra parts sometime in the future. Since I dont really have a need for a board that will accept 4 quad cores hopefully I can sell it and recoup most of the original cost on that. The case and power supply I will probably have a need for at some point, though they are much more than what the standard small business server needs. Anyone in Denver that happens to have a use for this stuff feel free to contact me.

The good news is that going to 1U instead of 2U will save the site $25 a month which we can put towards bandwidth if we need to.

[Simetrical]

Originally Posted by GrnEyedDvl

Sorry I havent been around much the last few days.

No problem, not all of us are college students on winter break.

Originally Posted by GrnEyedDvl

The board has been RMA'd to Tyan (Monday), which may take as many as 14 days. We cant really wait that long. We have to give a 30 day notice to our old host, and I dont want to gove them notice until the new machine is up and stable. That puts us up against the clock for the release of Empire. A 30 day notice for the Planet puts us in mid February, if I got the machine up and running today. Waiting a further 14 days puts us right at the end of February and I am not comfortable with that cushion.

Neither am I. March is when the game is released, so the demo is probably February, to judge by CA's past patterns (although in the past they've always released in the fall).

We have to give 30 days' notice before we move the server, but are we really going to put off starting the contract with the new host until that time is almost up? That will save a month's overlap in billing, but it puts us in a hard situation if there are problems with the new setup. Plus, it puts us on our current inferior hardware for an extra month. Or are you saying we'd be able to verify that it's working before the new host will start billing us?

Speaking of which, have you decided for sure which host we'll be using?

Originally Posted by GrnEyedDvl

Sooo... Today I ordered a 1U barebones system that will accept the 4 processors and RAM we already have. This system was not out when I ordered the original parts, damn them. It came out last week. Its a case, board, power supply, and rails with 4 hot swap drive bays at $1100. I am the only person that has ordered this from NewEgg, possibly the first person to order it ever. Since I have never had a problem with a Tyan board until building this machine, I am not too worried about more problems.

Because of this thread, I also ordered another 8 gigs of RAM, so we will have 16 total. That was another $230, but I want to eliminate as many problems as we can here.

All told I spent $1400 that I was not expecting to spend, and had it shipped next day air for $100. That should put it here Friday at the latest, possibly tomorrow if they get it out on time.

Yikes. Well, see the thread in Hex about trying out another ad provider. I don't deal with the money, so I don't have much more to say than that, I guess. 16 GB of RAM is probably no more overkill than 16 CPUs, I guess. Our database is now 13 GB and /var/userfiles is 18 GB.

Originally Posted by GrnEyedDvl

The good news is that going to 1U instead of 2U will save the site $25 a month which we can put towards bandwidth if we need to.

Didn't you say there might be problems with that size?

[GrnEyedDvl]

Originally Posted by Simetrical

We have to give 30 days' notice before we move the server, but are we really going to put off starting the contract with the new host until that time is almost up?

Probably not, but I dont want to run a full 30 days with both hosts unless there is a damn good reason for it. I was thinking 7-10 days overlap at the most.

Speaking of which, have you decided for sure which host we'll be using?

Probably Axis as discussed before unless someone else comes in with a better price. I need to talk to them again but I have been putting it off until I can give a firmer commitment on time. At that point I will hammer them on costs and see what I can get from them, and a few others.

Yikes. Well, see the thread in Hex about trying out another ad provider. I don't deal with the money, so I don't have much more to say than that, I guess. 16 GB of RAM is probably no more overkill than 16 CPUs, I guess. Our database is now 13 GB and /var/userfiles is 18 GB.

What I spent today doesnt change the amount the site needs to lay out, except maybe the extra RAM. I havent discussed that with Garb yet I just went ahead and ordered it. I will eat that if needed.

Didn't you say there might be problems with that size?

I was worried about heat, and still am a bit I guess. But if Tyan is confident enough to produce a barebones package for a 4 quad setup then its probably ok. Its actually a newer version of the same board we already have.

On the flip side of that, I could move all the stuff from the barebones case into the bigger case I already have if we ran into heat problems.