I sent this privately to Samuel and he suggested to me that I ought to be sending things like this to the list, hence this message. Samuel did respond to me concerning the issues I raised here, but I will leave it to him to decide whether he wants to forward his response to the list.
/Don Allen ---------- Forwarded message ---------- From: Donald Allen <[email protected]> Date: Sun, Aug 25, 2013 at 10:27 AM Subject: Re: Installing latest hurd To: Samuel Thibault <[email protected]> Samuel -- I got the contents of my home directory rsync-ed to the G41 running the Hurd. This brought up the first problem. This machine, while circa 2005, has a 3+ Ghz processor, 2Gb of memory and a 7200 rpm disk. While not a speed-demon by today's standards (the 3 Ghz is a bit deceiving, since it's an old Intel Xeon processor, which does not get as much work done per clock tick as the newer Intel processors; as I recall, the issue is that one of the pipelines is too deep, so a lot of work can get wasted when doing speculative processing of conditionals), its performance is perfectly adequate running Linux or Windows. I have run Linux on this machine for years (it was my primary machine for a good while when it was new) with an ext2 filesystem (I never used ext3, because ext2 was noticeably faster and Linux crashes were rare enough that the fsck time wasn't a problem; I also never lost a filesystem or a substantial portion therefore, which is theoretically possible with ext2, though I was and am diligent about backups). Transferring my 13 Gb home directory took hours and this was over 100 Gb ethernet. Doing the same transfer to a Linux system might take .5 hours, if that. The disk light was on pretty constantly on the G41, only occasionally on the source machine, which also has a 7200 rpm disk and is running Arch with a journal-less ext4 filesystem. My point is that it looks like the filesystem performance is very poor. Is there a buffer cache? Is it unified? The disk driver could also be an issue, depending on how it treats the disk cache. I managed to get X working (unfortunately, I did a text install, not pseudo-text; I wish I'd known the implications of that decision at install time, regarding the need to run the console command; I haven't reviewed the installation documentation, but if it doesn't discuss this issue -- it's possible that it does and I missed it -- it should). I don't use a desktop -- just a window manager (xmonad), dmenu, rox-filer, and something to provide info in the bar (xmobar). xmonad and xmobar were available in the repository and seem to work correctly. dmenu and rox-filer are not available. For this experiment, I managed without them. My first priority was a web browser. I installed iceweasel, which would not start. I got an error that talked about threads (I can be specific if you need it, though I suspect the project already knows that firefox/iceweasel doesn't work). I then tried installing midori, which did come up, but could not display any web pages, claiming it could not find the sites. Yes, I had entered the addresses of comcast's dns servers in /etc/resolv.conf and verified that dns was working by pinging a couple of sites. And again, anything I did, e.g., starting X and the window manager, starting midori, was painfully slow. Now it's true that the Hurd documentation does talk about the performance not being as good as Linux, but what I experienced is not usable, for me, and I doubt for many others. I would like to help with this -- performance analysis and improvement was a specialty of mine for many years -- but I am very busy with some things that I must do for my family. Hopefully this will settle down soon and if and when it does, perhaps I can devote some time to helping with Hurd. /Don -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected] Archive: http://lists.debian.org/CAPsTYbzVhBZM2h7=j8cdgrazwsa3thhdvuz2xyf6qdrndta...@mail.gmail.com
