| Wednesday, December 4, 2002 |
17:24 - Xserve Has the Way In
http://maccentral.macworld.com/news/0212/03.server.php
|
(top)  |
Mike sent me this link: a Network World story (reposted at MacCentral) covering the growing grass-roots movement toward OS X as a stable, simple, robust UNIX.
"Macs and management. Have you ever heard the expression 'like herding cats'?" asks Shane Wilson, coordinator of network services at Centre College in Danville, Ky. "Macs have always been this way."
The advent of Mac OS X 10.2, however, is changing that attitude. Network managers who formerly managed Macs with proprietary software and hardware can now use some of the same software they used for their Intel- or RISC-based servers and workstations to manage Macs. Mac OS X for the first time really supports standards-based enterprise qualities such as security, protocols and tools, which make management easier.
The article goes on to cite testimonials from IT managers of places like the H. Lee Moffitt Cancer Center in Tampa, Architectural Research Consultants in Albuquerque, and Anchor Group in Sacramento. Xserves are apparently starting to pop up everywhere, as people realize that they're extraordinarily easy to deploy and maintain-- as opposed to the typical steep learning curve and attendant high cost of expertise for UNIX/Linux servers-- as well as being far more flexible and powerful than the clumsy Windows servers that so many IT guys are wrestling unnecessarily with, sitting in server rooms poking at physical consoles and KVM switches instead of having any real remote access or peaceful coexistence of multiple services on a single machine.
Riddell uses the Mapics ERP application with Windows and Macintosh clients. He explains that even though he has not always been a fan of Macs in his network, he is softening with the advent of Mac OS X.
"Recently with Apple's OS X, I have completely changed my stance, and I am seriously considering the new XServe servers for a few minor Web projects," Riddell says.
On a related note-- I just sent in a message to John Polstra, author of CVSup, my favorite FreeBSD/UNIX mirroring application. CVSup's strength is the ability to track file creation, deletion, and modification, and propagate those changes very efficiently to the client machine (for backup or mirror purposes)-- using Rsync, it transmits only the crucial data, minimizing bandwidth usage. "Runner" processes zoom ahead and index the filesystems on both sides, communicating information about changed files back to the actual data-transfer processes, which transmit simple attribute modifications, appended data, RCS/CVS checkins, and finally (if all else fails) the entire file. This makes for an extremely fast and extremely bandwidth-stingy mirroring mechanism, and I love it.
There's one thing CVSup can't keep track of, however, and that is when a file is moved (or renamed). Because most operating systems, including UNIX and Windows, track files uniquely by their paths (e.g. /usr/local/www/data/index.html or C:\Program Files\Whatever\index.html), if you rename or move a file, the operating system has no idea of that file's history. As far as the OS is concerned, if you move /home/foo/1.txt to /home/bar/2.txt, it's a completely new file, and CVSup has to transfer the entire thing. It has no choice.
Not a big deal, you say? Well, what if I told you that I ran an art archive where each artist has a directory whose name is based on their displayed artist name. Let's say the artist has 1000 files in that directory, adding up to 100MB. Now let's say the artist changes her name. This means the name of the directory changes; but CVSup can't determine that the 1000 files under the new directory are really the same 1000 files that it had been tracking all this time under the old directory, and so it has to send delete commands for all the old files, and transfer all 100MB of the files all over again to put them in the new location.
Is that not the most inefficient thing you've ever heard? But that's the way it is on the operating systems that we all run our servers on.
But... the Mac's HFS+ filesystem has something I've mentioned once or twice before: the Unique File ID. The OS can track files not just by their paths (which is an ugly, frowned-upon, secondary "safety net" method), but by the Unique ID. This means that you can do such tricks as pointing an alias to a file, moving the target file around, and having the alias still work. Or put MP3s into your iTunes database, reorganize where you keep all said MP3s, and iTunes will still know where to find each one. Or configure your browser to find its "helper apps" in one folder under Applications-- and then decide to move the apps somewhere else entirely, and yet your browser will still work just fine.
And what's more-- if CVSup were retooled in an OS X build to take advantage of the Unique File IDs, it could track not only all the cool stuff it already knows about, but file renames and moves as well. And the above-mentioned scenario would result not in 100MB of extraneous network activity in the wee hours, but maybe 1K worth of control data that tells the CVSup client to simply rename the directory, as was done on the server.
(Sure, operating systems without the benefit of Unique File IDs might be able to employ some massively complex heuristics to determine whether a file with the same checksum and other attributes, but with a different path, is really the same as a file it had previously known about. But that is intensely inelegant-- we already have a better way, right here.)
So this could mean an Xserve would present a significant operational advantage not only over a Windows server, but over other UNIX platforms as well.
Sun thinks the Xserve is no competition at all. I dunno, methinks they could stand to quake in their boots just a tad more. Apple servers are no laughing matter anymore.
|
|
