On Fri, 23 Apr 2004, Andrew Cowie wrote: > People with Unix/Linux backgrounds are accustomed to administering > machines remotely and likewise used to doing things via command line > interfaces. But in a large in house corporate data centre, people > sometimes (for legacy reasons if nothing else) have the infrastructure > they need to KVM large numbers of machines.
Windows has had inbuilt remote access tools for years, using the PDR protocol from Terminal Services (no, you don't need a terminal Services license to use it). Its much better than most X compression (excepting the GPL stuff from nomachine.com) Many rackmount Windows boxes are shipped headless with the admin expected to log in through Remote Desktop to do anything on the machine. There's also a Web Interface (kidna like webmin) that comes with Windows 2003 Server and is valiable for Windows 2000 if you get the SAK version. Not advocating Windows (I work for Red Hat and there's plenty else I don't like about it) but thought there were better areas to compare than remote admin. > To the actual performance question: I imagine that the assumption is > that running an X server on a machine that "doesn't need it" is wasteful > of resources. Certainly if you're *using* that X server for a graphical > user environment, you'll be generating a lot of load that will detract > from the horsepower available for real services. How much load does X really generate? XFree86 might use say, 30M of RAM - that's not that much these days. Better open source X implementation (say, KDrive) use even less. It seems more common these days that people expect X to be installed on a server - even if its just to open a lot of terminals, rather than run GUI admin tools. > The assumption would seem to be that since a Windows server needs to run > it's graphical environment, whereas a headless (presumably implying X > -less) Linux box does not, a Windows server must under-perform a Linux > one. The weird thing is, the headless Windows boxes actually use a 'null' VGA driver. The GUI is welded in, but GUIs are just as remotable as the command line (depending on yoru bandwidth - but DSL is cheap). > I'm not so sure how much the GUI matters in reality. I think the real > key is "using the graphical interface". While X is a large beastie, to > be sure, if it's just sitting there not doing anything then a lot of it > can get swapped out - and will be as more active server processes > consume resources. Indeed. > Which all makes me think that a Windows server's GUI may not be as much > as a penalty performance wise as we might think. > > It would be interesting to find some unbiased statistics about all that. > Another domain, Java VM performance, shows Windows servers doing just > fine. From a year ago: > > http://www.volano.com/report/ Sun spend more time on the Windows JVM than the Linux or Solaris ones, cause thats what most of their customers use. According to some fellows I've met who've heard that from sun (trust that if you want, or don't - could be dodgy). > Doesn't mean I would recommend Windows by any stretch. I'm an operations > guy - the ability to remote administer farms of Linux systems and the > ability to finely control system layout, activity, and performance is > the critical differentiator. Me too. With Windows I often run into a brick wall when I'm troubleshooting - the only solution is to run some proprietary fixing tool or reinstall. With Linux I can drill down into the system to fix things. Its always good to hear from you Andrew, come grab a beer next time you're in Melbourne. Mike -- SLUG - Sydney Linux User's Group Mailing List - http://slug.org.au/ Subscription info and FAQs: http://slug.org.au/faq/mailinglists.html
