How to setup Debian XXen - a guide to building your own Xen hosted OpenSSI clustered machine.

"The OpenSSI project is a comprehensive clustering solution offering a full, highly available SSI environment for Linux."

"Goals for OpenSSI Clusters include availability, scalability and manageability, built from standard servers. Technology pieces include: membership, single root and single init, cluster filesystems and DLM, single process space and process migration, load leveling, single and shared IPC space, device space and networking space, and single management space."

Simply put, with OpenSSI, a group of physical machines "look like" a very large and powerful single machine.

CLVM is a Cluster enabled version of [[LVM]], using CMAN and DLM.

[[EVMS]] can be clustered using Linux-HA (heartbeat).

The syndication module in bloxsom is causing me some grief. Rather than fight with it at the moment, I've removed the entire right-hand column from my layout. I've also removed the Full Category Index at the bottom of every page. The static content should be much smaller now.

If you still find you're having problems, you can insert "/blog/" after blenke.com, but before the path you're looking for. This uses the dynamic page generation mode of bloxsom, which while slower, is much better about rendering things directly. By default, all pages (save those generated by the search function) are statically rendered, and should load immediately now.

Requirements:

• Integration with the Prius bluetooth handsfree.
• Addressbook sync with the Prius via obexpush (via Multisync, etc)
• Act as a wireless access point using bluetooth DUN uplink via Verizon 3G
• Act as a wardriving packet recorder at all times.
• Play audio via FM transmitter
• Potentially install the ViewTech unit to permit live DVD, video game (MAME, et al), and/or a live wardriving report ala kismet.

For integration with Vonage over a Verizon 3G phone:

• Get the CarPuter to act like a phone and talk to the Prius with Bluetooth HandsFreeProfile support for Asterisk, or ALSA support for BlueZ.

• Get the CarPuter to act like a handsfree for a bluetooth cellphone with Bluez Handsfree Emulation.

Ugh, this is so maddening. All I wanted to do was have a home VIA EPIA-M based system that could drive a TV for Myth while simultaneously driving a dual-screen desktop setup. You would think this would be easy (ok, I would think this would be easy).

Ideally, what I was looking for was a way to run two simultaneous X server instances. Display :0 would be the dual-screen nVidia Geforce2 MX/400 PCI card. Display :1 would be the onboard Unichrome VIA chipset driving the TV.

Unforunately, there exists real limitations on Linux Virtual Console support. Unless you're using kernel framebuffers for each video card, you only have the builtin VGA console. Spawning an X server allocates a VC from the VGA console driver, and you can only have one active at a time. Sure, you can spawn a single X server to drive all of the displays simultaneously, but this has a number of drawbacks if you wish to run a inputless PVR/MythTV display.

If you use kernel framebuffers, you run into other problems. The VIA Unichrome support, for example, destroys any hope of getting hardware MPEG or accelerated 3d graffics working. Likewise, the nVidia driver can have issues with the rivafb driver. And don't even think about using vesafb or vesafb-tng and getting both to work.

So I've done a bit more research and found that the linuxconsole project is mentioned quite thoroughly in the XFree Local Multi-User HOWTO. There is a "backstreet ruby" for 2.4 kernels which was backported during the 2.5 dev cycle, and a "ruby" 2.6 kernel patch which applies beyond those things that were integrated into the mainline kernel. The premise is to use the latest development tree ("ruby-2.6") against a 2.6.x kernel (see their CVS repository), and append the "dumbcon=X" option at boot time to allocate extra "dumb consoles" for X servers to use.

The first step was easy, if not CPU intensive: emerge an optimized Gentoo build, following various guides on the EPIA wiki.

The second step was to figure out how to "patch" a recent 2.6 kernel for this Ruby nonsense. Rather than a kernel patch, the project apparently expects you to copy a CVS checkout overtop your kernel build tree. Looking at the Makefile, it appears the CVS tree is a delta against 2.6.12.

cvs -d:pserver:anonymous@cvs.sourceforge.net:/cvsroot/linuxconsole login
cvs -d:pserver:anonymous@cvs.sourceforge.net:/cvsroot/linuxconsole co -P ruby-2.6
rsync -av --exclude=CVS ruby-2.6/ /usr/src/linux/

Now the kernel is building. Time for sleep. More on this soon.

The v9fs filesytem is an implementation of the Plan 9 network filesystem for Linux. It may soon become part of the 2.6.13 standard kernel tree.

It sounds like LANL is considering v9fs as part of their xcpu cluster job submission and general job management. Interesting.

The successor to the AVFS project, FUSE has made it back into the 2.4 and 2.6 kernel trees as a userspace harness for rapidly developing filesystems.

• Wayback is a versioning filesystem for Linux (think CVS or Subversion meets filesystem)
• Phonebook is a deniable steganographic encryption filesystem (multiple encryption keys, each presenting an entirely different filesystem)

Robert Hamburger, of RealUltimatePower.net fame, has published his book: Real Ultimate Power : The Official Ninja Book.

I don't know about you, but I'm PUMPED just thinking about it. Totally sweet.

Plan:

1. Get a Verizon PocketPC phone with unlimited 3G service for $45/month.
2. Install a SIP VoIP client on it.
3. Get Bluetooth networking working with Prius CarPuter
4. Potentially get Bluetooth handsfree working.

If you play with packaging systems long enough, you begin to see patterns.

• Slackware has tarballs.
• Debian has dpkg, and the wonderful apt dependency resolution tool.
• Redhat, SuSE, Mandrake, etc have RPM and various dependency resolution tools like YUM, apt for RPM, and others.
• Gentoo has portage/qpkg.
• Mac OS/X has the installer, OpenDarwin ports, and Fink
• Freebsd has ports and packages.
• Solaris has pkgtool
• HP-UX has sdux.
• AIX has LPP

etc, etc.

Moving back toward Unix think, it would be great to add metadata forks directly to files in the tree.

To find what package a file belongs to, you might check the blongs_to metadata fork for example.

To find what packages depend on a file to exist, you might check the requiredby metadata fork (requiredby={package},{package}). Each new package install would tag files it requires with its own package name listed in the resource.

Files might then have a "version" metadata fork. Binaries that have vanilla names (like "/bin/ls") would then have a reliable exposed interface for checking the version of that binary.

Packages themselves would still require a repository, primarily as an anchor for metadata about the package (package dependencies, conflicts, etc). I would argue that the "repository" be a directory structure of text files; each text file would list all of the files in a package (a quick forward index, negating the need to walk the entire filesystem to enumerate the files in a package).

It's just an idea, I'm sure others have thought of it, and it only works if you can tag files with metadata (or provide a userspace metadata overlay that isn't too obnoxious). This would require Yet Another Packaging System though, and I'm not sure the world really needs that.

Developed by Coraid, ATA Over Ethernet (AoE) has become a part of the core 2.6 Linux kernel.

There are two pieces to AoE, as there are with iSCSI. The initiator has only one implementation, that is included in recent Linux 2.6 kernels. The target has an implementation in the aoetools project named "vblade", and there is rumored to be another that Alan Cox once cobbled together.

The spec is open, however, so there is nothing stopping other implementations of AoE for other platforms to surface (in fact, I would expect to see at least a Windows implementation from Coraid at some point).

iSCSI - RFC3720 - SCSI tunnelled over IP packets.

There are two pieces to iSCSI, the initiator and the target.

An iSCSI initiator is the "client" that sees the drive on the host that mounts it as a filesystem.

An iSCSI target is the "server" that exposes a target drive.

There are now a couple of iSCSI target implementations for Linux:

• The iSCSI Enterprise Target Project (iscsitarget) (2.6 kernel) is a fork of the Ardis iSCSI target (2.4 kernel)

• The UNH iSCSI project has implemented a Generic SCSI Target Middle Level for Linux (SCST) to build upon.

Likewise, there are a number of iSCSI initator implementations:

• Open-iSCSI recently merged with the Linux-iSCSI(sfnet) project (often referred to as the "Cisco" initator), and appears to be actively in the middle of a major rewrite with performance in mind.

• The UNH-IOL iSCSI Consortium ("legendary in Fibre Channel circles") has the UNH-iSCSI SourceForge project with both a target and an initiator.

• Intel's iSCSI reference implementation has rather little following, but is available.

• UMass Lowell iSCSI hasn't been maintained for 4 years now, and really shouldn't be used for production purposes.

So I'm told to put together a network FAXing solution for a group of laptops running Windows XP (each with firewalls, that may or may not be at a home office).

What else better to use than Hylafax.

There is at least one great guide on configuring hylafax on a debian box, just look around. This is the easy part, really.

The problem is really with clients.

Option 1: http://winprinthylafax.sourceforge.net/

This option requires a client piece to be loaded that installs a printer driver to present the user with a FAX printer. When the user prints, a simple dialog box pops up asking for the phone number and email address to send the result of the FAX job to. When the FAX is sent, or errors out, the server will send a result email to the provided email address.

Simple. Almost too simple. No coverpages. No multiple FAX numbers. No contact addressbook to save FAX numbers to. Not much of anything. But easy to understand. Also: FREE ($0/seat)

Downside: requires client install.

Option 2: http://www.ifax.com/content/view/36/88/

HylaFSP. This option is like Option #1, but uses Microsoft's "Windows Fax Wizard" instead of the simple dialog box. This is commercial, $36/seat.

Fairly simple. Very user friendly. Actually supports multiple recipients and groups of FAX numbers, though we will need to test to see if it works with Lotus Notes' Addressbook.

Downside: requires client install.

Option 3: http://inconnu.isu.edu/~ink/new/projects/smbfax/

This is a completely different way of handling the print jobs. There is no client piece to load - the user merely maps a printer to the Samba server and prints their job as Postscript. The server then sends them an email with the URL to click on to access a web page for setting the phone number and coverpage to use when sending the FAX.

SMB Client -> Samba -> smbfax -> SMTP -> cgi-bin -> smbfax -> hylafax

Easy to support (nothing to load on the client), but perhaps detrimental to workflow for the user (they will need to check their email and click on a web link to finish the FAXing process everytime they send something to be FAXed).

Option 4: http://www.boerde.de/~horstf/

This uses a Samba print server like Option 3, with one change: instead of running a web server, there's a small Delphi GUI application that asks for the number to dial. The way this works is a bit different: the FAX server connects back to the client that printed the FAX on TCP port 5555 and the client machine fires off a GUI Application window asking for the number to dial.

This requires a hole to be punched in their XP firewall. In a closed environment, I would love to do this, but with laptops that are coming in over dialup connections.. not a good idea.

For more info on this, see Option 5.

Downside: requires client install. Upside: the client install is generic and simple.

Option 5: http://www.adixein.com/fax/

This is basically option 4 with a different GUI. It does a better job of explaining what is going on with Option 4, however.

Option 6: http://www.uli-eckhardt.de/whfc/

This starts the "thick client" method of talking to a hylafax server. This GUI is free, and provides the user with a bit more information than they really need for faxing things. Also, very simple and messy (raw tiffs and postscript files). Not something I like. The neat bit here is a FAX admin tool for viewing pending FAXes incoming and outgoing.

Option 7: http://www.cypheus.de/frmhomee.htm

Like Option 6, only commercial. Much easier to use. ~$30/seat.

For Mac OS/X workstations, there only seems to be one solution:

• A Java based Hylafax client

For Linux workstations, there are a few solutions:

• KHylafax - KDE based Hylafax client
• GFax - GNOME based Hylafax client
• A Java based Hylafax client

If you're looking for a FREE Global faxing system, check out tpc.int, a network of hylafax servers all over the planet that will gateway your faxes free of charge (as local calls).

First, two GPL'ed Broadcom "vendor" drivers, rather than the vanilla 2.4 kernel's asplodey bits: one for bcm4400, and the other for bcm5700.

Second, VMWare Workstation 5.0.0 build 13124 requires vmware-any-any-update91, and has a different vmware-tools-distrib set of kernel modules for those VMWare images I use for testing our new builds.

Thirdly, QEMU 7.0.0 now supports a VMWare like virtualized CPU kernel module. It looks like Fabrice isn't fully GPLing kqemu, but he deserves to make a living with all of his great work.

Things that need some followup:

• Opensourcing my "kerncob" harness for building debian standard kernels with all of the handbuild kernel modules and patches we use outside of debian proper.
• Starting a public apt repository along the lines of backports.org for those of us stuck maintaining large farms of Debian/woody servers.

6/14/2005 - Opening day for OpenSolaris.

Were you one of the first 5000 to grab a shirt?

With neato things like ZFS and zones, I'm downloading it now to play with it a bit under VMWare.

AdiumX is a wonderful multi-protocol instant messaging client, but it lacks Jabber MUC bookmarks. Someone has a creative AppleScript to create Adium Group Chat Bookmarks outside of the app.

Also a great guide toward scripting other Cocoa apps.

J-EAI is an XMPP-based Enterprise Application Integration (EAI) platform (also known as ESB, the Enterprise Service Bus). It is composed of several components, including an Erlang XMPP server core, connectors that support Open Adaptor and XSLT, and several distribution mechanisms, including publish and subscribe and content-based routing.

OpenAdaptor can be loosely classified as EAI (Enterprise Application Integration) software. It is highly extensible and provides many ready-built interface components for JMS, LDAP, Mail, MQ Series, Oracle, Sybase and MSSQL Server as well as data exchange formats such as XML. New components are regularly added.