|
SUSE GOES ONE UP WITH WIRELESS DESKTOP
While the hype centers on 3G
Internet phones and Bluetooth PANs, |
![]() |
|||
by
Jack Fegreus |
|
Could Windows be dead on the desktop? That question is far less fanciful than it might at first sound. While the financial analysts all continue to lionize Microsoft as the shining knight of IT investment, a number of chinks are starting to show up in the armor. Certainly as a trademark, the Windows™ gold may turn to dross as the developers of LindowsOS have decided on a full legal challenge to the Windows mark. Such a courtroom brawl will at a minimum serve to catch the attention of the general public. What better time to capture the public’s attention, At least on the server side, it is now a two-horse race between Linux and Windows 2000. Better yet, many market analysts now voice the opinion that Linux is ahead by a nose in the race to install new IT servers. |
|
While technically just a “point release,” SuSE 8.1 provides plenty of improvements for both IT and end users. For IT, there is support for touch screens, systems with multiple monitors, and the ability for a computer to have multiple system profiles on a network. There is also support for the Apache 2.0 web server, an improved YaST2 runlevel editor to start and stop system services, and integration of the YaST Online Update (YOU) module into package management. In addition, the new distribution includes the XFS file system from Silicon Graphics, which means there are now four journaling file systems from which to choose: ReiserFS, Ext3, JFS, and XFS. Nonetheless, the most interesting features of version 8.1 intersect the needs of IT and those of corporate end user. These features center on SuSE 8.0 Pro Office, which included version 6.0 of Star Office. When we reviewed SuSE 8.0 with Star Office 6.0 and Evolution 1.0, openBench Labs focused on the feasibility of a typical knowledge worker running Linux on the desktop. That meant assessing the ease-of-use and the transparency of that software troika vis à vis running MS Office on either Windows XP Pro or Windows 2000 Pro. |
|
We were primarily interested in three things: How well could StarOffice 6.0 deal with complex documents from MS Office, how intuitive was the user interface for someone who regularly used MS Office, and how well could we share our ersatz MS Office documents with other users on both PCs and Macs? On all three points, StarOffice 6.0 was an outstanding success. For the first time testing an MS Office-compatible suite, we encountered no problems opening Office 2000 or Office XP documents. In the 8.1 release of SuSE, OpenOffice 1.0.1 replaces Star Office 6.0. OpenOffice is Star Office without any of its software modules that include proprietary code. While it’s a bit of an oversimplification, for practical purposes you can think of OpenOffice as Star Office without ADAbase. With several Open Source databases to choose from on the SuSE distribution, the loss of ADAbase in OpenOffice is hardly a matter for concern. We concluded our previous look at version 8.0 by investigating the ability of Linux to communicate and utilize USB-based devices such as a digital camera. We tested USB support by plugging an Olympus 4040Z into the USB port. Using Kdiskfree on both the GNOME and KDE desktops, we were able to dynamically mount the camera as a new storage device and navigate its directory tree to manipulate the photos stored on its Smart Media card. SuSE 8.1 extends this support by adding the ability to hot plug USB2 and Firewire devices. |
|
The two biggest changes, however, in the new release are the inclusion of the GNOME 2 desktop environment and the addition of built-in support for wireless LANs. Of these two, GNOME 2 is the most problematic in an OS-upgrade scenario. For openBench Labs, each system upgrade that we made from SuSE 8.0 to SuSE 8.1 required a two-step approach. The reason for this is rooted in the SuSE YaST installation program. It views all of the GNOME 2 packages as new software rather than as upgrades to GNOME, which is technically correct. As a result, a SuSE 8.0 system that has GNOME as a desktop option will not have that option after upgrading to version 8.1 of SuSE. Nonetheless, all of the appropriate GNOME 2 modules can be easily added via YaST following the initial shock of booting into KDE. |
|
|
Once GNOME 2 is correctly installed, the default window manager is Metacity and not Sawtooth 2. Metacity is touted as being a “lighter weight windows manager.” Certainly, this is true in terms of configuration options. For those coming from Windows, the rather Spartan configuration options will feel quite natural. As for Metacity’s performance, we measured no significant difference between it and Sawtooth on GNOME 1.4. As previously measured with version 8.0 of SuSE, applications ran in the range of 3% to 5% faster on GNOME 2 than KDE 3.0. The real distinguishing feature of SuSE 8.1, however, is its extensive wireless LAN support. Wireless LANs are not really new to Linux, as openBench Labs tested the XGate Linux-based wireless LAN server a year ago. The significant difference is that the XGate was explicitly configured to work as an Access Point with a packaged Prism PCMCIA card and nothing else. Such pre-cooked connectivity is of little use with today’s new generation of super-slim laptops, which have created a new twist on the connectivity issue: how to stay connected in the office when walking around. Mobility began to emerge as an issue with the need to make slide presentations in meetings. Next, there was the need to crunch spreadsheets on the fly. But just as it became handier and handier to disconnect a laptop from the office LAN and bring it along to a meeting, dependency on email and the web made it harder and harder to disconnect for any extended period. Moreover, that sort of connectivity issue is not the kind of problem that a 1-Mbit Bluetooth-based PAN is going to solve. On the other hand, going back to the future with an 11Mbits per second LAN that extends my 100Mbits per second wired LAN and leaves me unrestrained to walk around is an easy price to pay. At the heart of this technology is the wireless Access Point (AP), which forms a bridge between the wireless and wired LAN. The AP is analogous to a base station used in a cellular phone network. With one or more APs present in what is dubbed “infrastructure mode,” individual clients only communicate via an AP and not on a peer-to-peer basis. What’s more, mobile clients can roam between APs, which makes seamless campus-wide coverage possible. |
|
While there has been a welling tide of wireless LAN devices for Windows and Macs, up to now there has been a dearth of modular drivers to enable these devices to work on systems running Linux. On Linux, going wireless has, more often than not, required a recompile of the OS—not a big deal for enthusiasts, but seriously frowned upon by corporate IT. So in corporate IT, wireless clients run Windows, while more and more of the services these clients access run on Linux. That’s why in all of our previous openBench Labs wireless networking tests, we used laptops running Windows to access servers running Linux. This time we were able to reverse that interoperability scenario. Using SuSE 8.1 on a laptop, we accessed data on a Windows 2000 server over a wireless LAN with remarkable ease. |
|
|
To insure maximum interoperability with other wireless LAN interface cards and APs, it’s best to look for devices bearing the Wi-Fi mark. This indicates that the equipment has successfully passed the interoperability tests devised by the wireless industry group. We chose the Proxim ORiNOCO AP 1000 for our access point. The AP-1000 utilizes two PC Card slots to enable transmission on two different channels for load balancing. Currently there are two ORiNOCO PC cards available with either the typical 64-bit Wired Equivalent Privacy (WEP) algorithm from RSA Data Security or a more robust 128-bit RSA encryption algorithm. We used 64-bit WEP cards in both our AP and our HP Omnibook client. For anyone trying to integrate Linux and Windows 2000 into a corporate file-sharing scheme, it is either necessary to make Linux dance to the tune of Windows using SMB, or put Windows 2000 in step with an open industry standard protocol. This boils down to putting Samba on Linux or NFS on Windows 2000. For environments with casual integration needs, Samba lets Linux clients cut in without missing a beat. It’s when you introduce a few more operating systems, say Solaris, HP-UX, or maybe even OpenVMS; you require a more exacting distributed security system; or you start to build an enterprise IT architecture on top of a SAN, that the Windows Samba solo starts sounding a bit discordant. Just consider the minefield of file security. Linux file systems apply ownership and security from a completely different perspective from Windows 2000. Under Linux, only individual users own files. Under Windows NT/2000, individuals and the groups to which they belong may own files. For NTFS, Windows 2000 begins applying security from a global perspective—Everyone— and then begins to look membership in subset groups. Linux, on the other hand, starts with the user’s UID, then turns to the group to which the user belongs via the GID, and then finally looks to world permissions. |
|
DiskShare provides a number of security mechanisms to rationalize the Unix security scheme with the NT security scheme. At the least granular level, there are global share permissions for resource—disk volume—access: none, read-only, read-write, and root. At the file level, read, write, and execute permissions are based on user and group IDs. As a result, DiskShare has robust facilities to map any given user ID to any Windows NT user and any given group ID to and Windows NT group. UIDs and GIDs can be entered manually or imported from a Network Information Service (NIS) master server. Once DiskShare is up, running, and properly configured by a network or systems administrator, it creates an NFS daemon process on the Windows server, dubbed the PCNSFD service, which enables users to share any disk, partition, or folder as an NFS volume just as they would share a volume via SMB. The volume that we chose to share was physically resident on our MegaRAM 2000 ram disk, which is fully capable of saturating our 1Gbit SAN. We chose such functional overkill in order to insure that the wireless networking tests were not adversely affected by ancillary bottlenecks. |
|
To test performance we began by creating a performance baseline for wireless file transfers using Samba. Our tests consisted of simple simple file transfers of a 100MB file initiated by dragging and dropping the file using the GNOME 2 Metacity desktop manager. We monitored the laptop system's performance using a KDE application, ksystemguard. GNOME 2 is perfectly capable of running applications with GUIs designed for KDE and GNOME. We used Linux Network Neighborhood as our client Samba application. This application provides a very comfortable interface for corporate Windows users. Interestingly, the latest version of this Samba client has eliminated the option to scan the network using a specific range of IP addresses. This change in the client application was transparent when using the Omnibook's 3Com embedded LAN controller; however, problems arose when we tried to find systems with our wireless connection. |
|
|
Scanning our workgroup using SMB using a wireless connection, however, never returned any of the available Windows computers or Samba servers. Only by explicitly adding a server via its IP address could we mount a shared volume offered by that server. Because of this wireless scanning issue, we recreated all of our tests with a Fast Ethernet connection. All of the results scaled according to Hoyle. In tests where we measured throughput in the range of 400KB with a wireless connection, throughput jumped to 4MB per second. Whatever its cause, the SMB scanning problem over a wireless connection did not materially affect performance once a connection was established. When copying data from the Windows server to the laptop via Samba, throughput peaked at around 450MB per second and averaged about 375KB per second. At the same time, a steady 10% user process load was imposed on the CPU. |
|
|
When copying our test file to the server, via Samba, throughput was a bit more consistent. This time it peaked again around 450MB per second, but was able to average around 425MB per second. Once again, CPU overhead had a distinct user process component of about 10%. Shifting to using NFS on the client and using the PCNFSD daemon that was created on the Windows server with DiskShare, we measured a very distinct improvement in performance. We started by doing a file copy to the laptop from the same volume on the server used in our Samba test. This time, however, the volume was mounted via NFS. Throughput jumped by 70% to 635MB per second. At the same time, the CPU load was characterized by only a small amount of initial user process activity and a corresponding increase in the system load activity. The net result was a file transfer that completed in almost half the time. |
|
|
Copying a file to the server using NFS did not produce any statistically different results from copying from the server. Throughput averaged about 650MB per second, which translates into about a 45% improvement in throughput performance over Samba. Once again, CPU overhead was a function of system process load rather than user processes. In terms of the time necessary to make the file transfer, NFS was about 30% quicker. As noted earlier, when we repeated these tests over a wired Fast Ethernet LAN the relative results were unchanged. Using a medium that was an order of magnitude faster simply raised all of the performance numbers accordingly. The first inescapable
result is that using NFS v3 provides a significant throughput advantage
sharing files. The second is that despite all the hand-wringing over the
folding of Eazel earlier this year, Linux is very much alive and well on the
desktop. |
|