CUNIX Hosts

Below are the hosts that comprise the CUNIX cluster. Most of them are not available for users to log in to directly, but serve various backend functions. Here is an old list of each host with its configuration (model, operating system, memory, peripherals, etc.).

cunix
mail, news CourseWorks ColumbiaNet files timeshares authentication & authorization administrative
popimap
cubmail
sendmail
news
CourseWorks front-end
CourseWorks database
web hosts
legacy
file servers
tape backups
login hosts
cpu servers
kerberos
LDAP
tacacs/tacplus
ID system
database backend
DNS

Mail Servers

$ ourhosts -mail
maillist1 mhoro blueberry cranberry elderberry hello manheru merdeka
moni morjens raspberry strawberry tere xinchao zravo apakabar kachifo
menyapa
$ ourhosts -popimap
$ ourhosts -cubmail
$ ourhosts -maildeliver
When
a user reads mail, the mail user agent (e.g. Netscape, Outlook) contacts a daemon on the popimap machines to retrieve their messages. While pine on the timeshare hosts accesses mail directly (over NFS), POP and IMAP requests can come from any machine on the network. Hosts should be configured to connect to imap.columbia.edu or pop.columbia.edu.

Cubmail, a convenient, secure, web email client, is increasingly popular with students. The cubmail front-end hosts connect to a database backend to track session information and address entries. Currently, the cubmail hosts use a separate pool of IMAP servers, which may eventually be joined with the rest of the POP/IMAP servers.

The sendmail hosts are further in the background -- they receive incoming mail addressed to user@columbia.edu, and deliver it to the user's mail spool or relay it to other hosts at Columbia. They also expand mailing lists.

In 1999 we made plans (available internally) to consolidate, upgrade, and simplify our mail system. Except for the usual problems keeping up with demand, those plans have mostly been accomplished.

News Servers

$ ourhosts -news
newsmaster
AcIS currently has one host handling
netnews -- news.cc.columbia.edu or news. It is the host news readers connect to, and should also be used for posting news. The same host acts as our newsmaster; it is the host contacted by non-Columbia sites that are passing news to us. The newsmaster sorts through duplicate messages from multiple feeds and organizes them all into a single collection for Columbia hosts.

Given that netnews is the single largest consumer of campus network bandwidth, the system is surprisingly stable. Other Universities have given up maintaining their own news systems in favor of commercial providers.

CourseWorks

$ ourhosts -cmsent
gooseberry huckleberry loganberry
CourseWorks@Columbia, a Course Management System using
Prometheus, is not yet in production. Internal planning documentation is available here. It will have two front-end webservers and a backend database.

Web Hosts

$ ourhosts -cuweb
alhan howsit bomdia gutentag kalimera moningtru zivijo osiyou yasu thalia
$ ourhosts -www
$ ourhosts -http
$ ourhosts -https
The web hosts provide a number of specialized functions.

Recently, the main secure web server was merged into the same pool as the regular web servers. Merging the two simplified maintenance and increased redundancy for the secure web (which had previously been dependent on just one (older) host). The combined cluster provides most of our authenticated services, like bulletin boards and internal documentation. Some of the authenticated services, like the backend of our account maintenance system run on other secure servers.

Another important web service is our various search engines, which require large indexing jobs to privide up-to-date answers. Currently, they all run on one host.

We also have a Certificates on Demand (cod) server, an authentication broker (wdamb) and a commercial server for the restricted portions of Earthscape.

Here are some more details, slightly dated, of the web hosts, their functions, and the plans for upgrading them.

ColumbiaNet hosts

$ ourhosts -cnet
$ ourhosts -cnetsunset
The original campus-wide information system was written with (or like) gopher and was text-based. ColumbiaNet terminals on serial lines were placed in various locations around campus for public access. ColumbiaNet is also available for users to telnet into from both on- and off- campus. The system involves a number of inter-dependent programs running on the ColumbiaNet hosts.

The information contained in ColumbiaNet is gradually being moved to the Web, but some services still remain available via ColumbiaNet. For the most part, these are text-based alternatives to services available on the web, some of which are better or more stable than the web versions.

Many of the services on ColumbiaNet include database lookups and user authentication. These take place on the ColumbiaNet back-end.

The ColumbiaNet front-end and back-end servers are still running SunOS. Fortunately, they are fairly stable and we rarely have to do anything to keep them running.

Still, we have resigned ourselves to the fact that ColumbiaNet will not be vanishing soon, and have finally ported core services to a Solaris host, informally known as "ColumbiaNet Sunset". This will be deployed shortly.

File Servers

$ ourhosts -toasters
cheers ellohay chinchin prost salud
$ ourhosts -toastersmp
$ ourhosts -toastmaster
Files for cunix users are stored on file servers and accessed via the Network File System (NFS). Thus, wherever a user ends up, they see the same home directory. This includes the timeshares, the Web servers, and even the Mudd Sun lab.

Similarly, incoming mail (also known as new mail, or mail spool) is stored on another file server, and Web and EDS data (/www, /wwws, and /eds) are stored on two more file servers.

Our fileservice is provided by Network Appliances filers, which the company used to call toasters. Unlike the general-purpose UNIX boxes we use for our other servers, a toaster does just one thing, but it does it really well. The NetApp operating system has been tuned for excellent file service.

The toasters are almost named for greetings, like our other machines, but we started with a greeting ("cheers!") which is also a toast, and continued naming subsequent filers after toasts as well.

Tape Backup Hosts

$ ourhosts -backup
$ ourhosts -veritas
kiaora sawubona
As with all computers, our disks need to be backed up. This includes the fileservers as well as the other hosts, whose disks contain important configuration information and system logs. These backups are done to DLT (Digital Linear) tapes. The tapes are stored in a large tape jukebox from ADIC, with a robot arm to load the right tape for unattended backups. The backup hosts take care of loading the right tapes, and signal each host in turn to run its backups. As many as 7 or 8 backups can run simultaneously. These generally run overnight.

Timeshares

$ ourhosts -t
aloha ciao konichiwa sawasdee merhaba
$ ourhosts -timeshare
$ ourhosts -login
To log in to the AcIS machines, users should connect to the hostname
cunix.cc.columbia.edu (within Columbia, this can usually be called simply cunix). When users connect to cunix, network software automatically chooses one of the timesharing machines, and they are connected there. An individual timeshare machine gives them the login prompt. The timeshares are all identically configured.

The timeshares have been expanded many times since their initial deployment in 1993. At first, the cluster contained four SPARCstation 2s. In 1994, we had five SPARCstation 10s. In January 1998, we expanded from twelve SPARCstation 20s to six Sun Ultra 2s (each at over 3 times the capacity). Currently, we only have five Ultra 2s.

Plans call for upgrading the operating system but not the hardware. We are counting on decreased demand as more users do their work offline (e.g. pine users moving to IMAP clients).

CPU Servers

$ ourhosts -cpu
fozimane monire labdien
$ ourhosts -cpusolfhs
$ ourhosts -login
Some users of cunix wish to run CPU- and memory-intensive programs like SAS and lisp. In order to shield other users of cunix from the spikes these processes would cause, these programs are shunted to alternate CPU server hosts. The programs are configured to run on the CPU servers transparently, as if they were still on the timeshares.

Currently, we have two types of CPU servers, the new Solaris 2.7 one, not fully deployed, and the old Solaris 2.5.1 ones. Among other things, the newer one runs java, SAS and SPSS. The older ones run lisp, mathematica, and matlab (as controlled by /opt/local/etc/spty.conf).

Kerberos

$ ourhosts -kerb
tegur terve
To help combat the problems of plaintext passwords crossing a possibly-compromised network, MIT wrote a package called Kerberos. This gives out tickets based on authentication of encrypted strings. As with other services, AcIS has one main kerberos server and one backup server. As of late December 1999 we are running Kerberos 5 servers, with some Kerberos 4 compatibility for our many Kerberos 4 clients. While kerberized telnet is being replaced with ssh, sshd still uses Kerberos to verify the password.

Tacacs

$ ourhosts -tacacsrvr
cumulus stratus
$ ourhosts -radius
$ ourhosts -tacacs
To allow users to run SLIP and PPP from our terminal servers, while still being able to restrict who can do this and track who was on when, we need to authenticate users on the terminal servers by having them login. This is done by the terminal servers with the radius and TACPLUS (and TACACS) protocols, by contacting the corresponding daemon running on our radius and tacacs hosts (with backup servers on various other hosts). These servers in turn contact the directory name servers (for authorization) as well as the kerberos servers (for authentication).

LDAP

$ ourhosts -ldappool
ave hola yahteh
$ ourhosts -ldap
$ ourhosts -ldapmaster
For user authorization and simply looking up phone numbers, Columbia provides directory services (i.e. names of people), done with the LDAP package. We currently have three dedicated LDAP servers, one of which is the master LDAP server. For better performance, some hosts run their own LDAP servers for themselves and nearby hosts. The LDAP authorization service is used by systems like secure web and tacacs (for dialin IP) to verify a person's eligibility for different services on campus. For LDAP service, connect to host ldap.columbia.edu.

ID System

$ ourhosts -sidmaster
dobryden
$ ourhosts -sid
Cunix currently serves over 50,000 users, and maintains information on over 90,000 people (not counting alumni). In order to allow access for these users in a timely fashion, we utilize a database of all Columbia staff, students, and affiliates. This database comes from various University sources, is merged together, and is then used to verify potential users. There are then more database records created with information about each user. These records are then combined with records referencing over 90 hosts, and user entries are made available to the right hosts.

Newer Solaris 2.7 hosts, as well as authenticateed web services, depend on LDAP to get the user information, but LDAP ultimately gets the information from the ID system. The ID system is most visible at our Computing Support Center, where people go for help creating or renewing their accounts, but is constantly active keeping passwords up to date, and creating new accounts on demand.

Database Backends

$ ourhosts -ingres
bienvenue
$ ourhosts -db2
$ ourhosts -acscsun
$ ourhosts -cmsent
Our ID system, as well as our hosts database, class scheduling software, and a number of other systems, require a large relational database system. We use Ingres, and for optimal performance the database runs on a dedicated database backend host. Client programs run on many other hosts updating their various databases.

We also have some databases in DB2, with a backup server on the way. DB2 is primarily used for databases of web information.

Additionally, our Remedy tracking system uses Sybase, and our CourseWorks system uses Oracle. And postgresql for the CubMail database.

Name Service

$ ourhosts -named1
diaduit namonamah nihao labdien cunixd
$ ourhosts -named
Hosts on the Internet have a hostname and a number (IP address). With so many hosts at Columbia and elsewhere, no one host can keep track of them all, so each domain is responsible for helping other domains resolve local hostnames. The package used for this is known as DNS. For Columbia, several hosts are designated to keep track of the local hostname to number mapping, with four hosts currently being the main ones along with two or three backup hosts. The name service system also allows for dynamic mapping of hostnames -- so that subsequent translations of the name cunix cycle through the various IP addresses (via a round-robin list), though some of these pools are being moved to Server Load Balancing (SLB) on the routers.

Please check with hostmaster@columbia.edu if you need to know what address to use for resolving hostnames.


Last modified Oct 2 2001