On Wed, 5 Apr 2000, Lenz Grimmer wrote:

On Wed, 5 Apr 2000, Reid, Rowan (GSP) wrote:

...

with the advent of XFS on linux which will be the most stable and fast option. maybe i'm biased but XFS is -proven tecnology

But not on Linux :)

XFS has some interesting features, too. But it will need some more time, before you can really declare it as "ready for productive use". Our internal tests have ended in kernel freezes a lot of times. But it looks very promising! I did not expect SGI to get it ready that fast.

...and if it's ready, we'll be the first to include it as well. We have a great relationship with SGI, and currently have some people working their (helping with the port of SGI's failsave product, for High Availability, to Linux (OpenSource, i.e. GPL)). -- Michael Hasenstein http://www.suse.de/~mha/ Private Pilot (ASEL+aerobatics) since 1998 -- To unsubscribe send e-mail to suse-linux-e-unsubscribe@suse.com For additional commands send e-mail to suse-linux-e-help@suse.com Also check the FAQ at http://www.suse.com/Support/Doku/FAQ/

On Thu, 6 Apr 2000, Wayne Chan wrote:

Hi,

You've mention HA. Great! I am currently looking around for HA implementation, GPL softwares if possible. So far, LinuxVirtualServer.org looks promising AFAIK.

So when is SuSE coming out with their version of HA/load balanced? Any beta testing?

There are several solutions in one sentence here. Linux VirtualServer is already a good solution for webservers. The SGI HA stuff aims higher. We'll use it, for example, for Oracle servers, or fileservers, or... etc. You can already use the virt.serv. stuff. The SGI Failsafe solution (that's the name) needs more work. It's all user space, 650.000 lines of code plus Java GUI, and relatively easy to port and pretty clean, but HA is a complicated topic, so that will need a lot of testing before it can be used as 'production quality'. After all, this is supposed to be for big enterprise solutions, and not small department servers, and who wants to take blame when such a HA system fails because it wasn't fully tested? SGI Failsafe is a proven solution with a long history, but on Linux it is brand new. Beta-testing may be possible in a couple of weeks already. Look at www.linux-ha.org, the code is supposed to be integrated into the Linux HA project. All big Linux distributors are involved, we only started it. -- Michael Hasenstein http://www.suse.de/~mha/ Private Pilot (ASEL+aerobatics) since 1998 -- To unsubscribe send e-mail to suse-linux-e-unsubscribe@suse.com For additional commands send e-mail to suse-linux-e-help@suse.com Also check the FAQ at http://www.suse.com/Support/Doku/FAQ/

On Wed, 05 Apr 2000, Avi Schwartz wrote:

Lenz, I need your insight here. I am having problems running Corel WordPerfect Office 2000 on my laptop (the applications crash upon launching them) but no problem at all on my desktop machine. Now the kick is that the laptop uses Reiserfs while the desktop is still ext2. Tonight I met One of Corel's senior developers, Gavriel State, in the Corel Linux Roadshow 2000, an he suspects that there might be my problem. According to Gavriel, they use memory mapping extensively to launch their WPO2K applications and if Reiserfs has some incompatibility in this area it might cause my problems. Do you know if there is any problem with Reiserfs' memory mapping? BTW, I have no clue as to what memory mapping means.

Well, I don't know much about reiserfs (except that, from what I've heard, it's closer to being ready for prime time than either of the two lead alternatives) but I can tell you the last one. Memory mapping is a variation on virtual memory. Normal virtual memory starts with the real memory occupied and the swap space vacant - and, further, no correspondence between address space and swap space. Such correspondences are created when pages are swapped out, and can be destroyed when pages are swapped in. With memory mapping, a particular file is declared to be part of the swap space for a particular task. It is assigned a correspondence to particular pages in the task's address space - and the correspondence is permanent until the mapping is broken. And the swap space is defined as being occupied.

From there, in use it acts just like virtual memory. Most likely the next thing you do is look at the file's headers - which you do by means of memory access, NOT file i/o. Since the first page of the file isn't in RAM, you get a page fault, and the virtual memory manager goes and fetches it for you. The process is invisible to the application; the OS handles it. Similarly, any other pages the application actually needs are brought in, and pages that haven't been referenced in a while may be swapped out. Except that, so long as they are in the defined address space of the file, they go to the file, not to normal swap space.

Eventually, the mapping is broken: you close the file. At this point it is necessary to write out any dirty pages in real RAM, and possibly account for changes in file size (in both directions). The memory-mapping routines take care of this, not the application. -- To unsubscribe send e-mail to suse-linux-e-unsubscribe@suse.com For additional commands send e-mail to suse-linux-e-help@suse.com Also check the FAQ at http://www.suse.com/Support/Doku/FAQ/