**** BEGIN LOGGING AT Sun Apr 29 02:59:56 2007 Apr 29 03:30:31 03osas * r5992 10optware/trunk/make/asterisk14-core-sounds-en-ulaw.mk: asterisk14-core-sounds-en-ulaw: 1.4.5 -> 1.4.6 Apr 29 03:31:37 03osas * r5993 10optware/trunk/make/asterisk14-extra-sounds-en-ulaw.mk: asterisk14-extra-sounds-en-ulaw: 1.4.4 -> 1.4.5 Apr 29 03:32:05 03osas * r5994 10optware/trunk/make/asterisk14-extra-sounds-en-gsm.mk: asterisk14-extra-sounds-en-gsm: 1.4.4 -> 1.4.5 Apr 29 03:33:57 03osas * r5995 10optware/trunk/make/asterisk14-core-sounds-en-gsm.mk: asterisk14-core-sounds-en-gsm: new package Apr 29 04:09:10 03osas * r5996 10optware/trunk/make/asterisk14.mk: asterisk14: reducing the size of the package (we have a separate package for core-sounds-en-gsm) Apr 29 10:55:04 anyone here using asterisk on their slug? Apr 29 12:08:33 03osas * r5997 10optware/trunk/Makefile: asterisk14-core-sounds-en-gsm: promoting asterisk14-core-sounds-en-gsm package Apr 29 22:21:40 a plus tard... Apr 29 22:21:47 oops sorry Apr 29 22:22:21 anybody built HEAD since friday? mine failed on npe.. Apr 29 22:25:49 rwhitby has upgraded the microcode afaik Apr 29 22:25:53 haven't tried building for ixp Apr 29 22:26:41 it's never to late.... Apr 29 22:27:39 hehe Apr 29 22:27:52 ;) Apr 29 22:27:52 Got a wide range of powerpc devices I'm working on atm. Apr 29 22:36:30 I just know that make slugosbe-image failed. Apr 29 22:37:20 actually.... Apr 29 22:37:49 I just noticed that the npe package needs to be downloaded manually... Apr 29 22:38:11 Odd. I thought were weren't using intel's stuff anymore.... Apr 29 22:42:02 VoodooZ: we can't get away from the microcode Apr 29 22:42:09 we don't use the access library any more. Apr 29 22:46:52 I see. Apr 29 22:47:02 I've downloaded the latest and building.. thanks. Apr 29 22:47:39 I haven't touched the robot/slug stuff at all for a while. I have to start soon though... Apr 29 22:47:46 later Apr 29 22:51:05 rwhitby: btw, since we're not using apex, why aren't we using the upx compression that john reiser worked on? Apr 29 22:51:45 I'm discussing upx compression for a minimal kernel for powerpc right now, and that reduced the kernel from 1.4mb to 1mb (!) Apr 29 22:53:56 (Still in development, so it's not actually working yet) Apr 29 22:54:26 the memory usage is largely unaffected, since it uses in-place decompression Apr 29 22:55:46 sounds good to me - what's involved in supporting it? kernel mods? Apr 29 22:56:35 The kernel needs a new decompressor, so I guess it's a patch. Apr 29 22:56:43 it also needs some tools available when building Apr 29 22:56:50 ie, the compressor tools. Apr 29 22:57:11 does assembler code need to be written? Apr 29 22:57:24 afaik, it works for ARM already. Apr 29 22:57:39 so I don't think any new code needs to be written Apr 29 22:58:30 are the patches available? Apr 29 22:58:43 I compressed the userspace binaries in my em-mode rootfs. The binaries generally compressed to about 50% their original size. Apr 29 22:59:19 wow, it would work on initramfs too? Apr 29 22:59:32 I could definitely use this on ppc Apr 29 22:59:36 can't see any reason for it not to work. Apr 29 23:00:15 I assume there's a longer decompression time Apr 29 23:00:36 but even a couple of seconds would be worth it Apr 29 23:00:38 I think he mentioned that. I don't think the decompression time was that much longer. Apr 29 23:00:42 indeed Apr 29 23:01:43 I wonder if it'd be possible to get openembedded to automatically compress binaries when building. Apr 29 23:02:16 busybox compressed to 1/3rd of the original size on my fs. Apr 29 23:04:06 (Given, it's a tiny busybox to begin with) Apr 29 23:04:09 the board I have been working on has 4MB flash, now I'm trying to put linux on a similar board which only has 2MB flash Apr 29 23:04:16 a semi-default busybox for powerpc: Apr 29 23:04:17 850488 -> 479756 56.41% linux/ElfPPC busybox Apr 29 23:04:28 orig -> compressed Apr 29 23:04:46 my current kernel + initramfs image is 3MB Including some stuff I can remove, like bash ;-) Apr 29 23:04:48 that's got most of the utilities enabled. Apr 29 23:04:55 bash is ... huge! :) Apr 29 23:05:01 I know Apr 29 23:05:06 600kB+ Apr 29 23:05:10 and it requires ncurses, doesn't it? Apr 29 23:05:18 yep, the way I built it it does Apr 29 23:07:32 is that a 2.6-kernel btw? Apr 29 23:08:12 yep, currently 2.6.19 because it works and has proven stable Apr 29 23:08:24 how big is the kernel? Apr 29 23:08:38 I'm really struggling to get a 2.6-kernel down to an acceptable size Apr 29 23:08:56 A bit of the problem is that I have to have support for a couple of filesystems. Apr 29 23:09:06 but even without that, it's a bit large. Apr 29 23:09:13 I think it's right around 1MB but I'm not sure Apr 29 23:09:51 do you have to have ext3 builtin? that's big Apr 29 23:10:06 yes :-\ Apr 29 23:10:10 The worst bit is xfs Apr 29 23:10:16 with initramfs, I don't need *any* disk based filesystems Apr 29 23:10:25 one of the benefits over initrd Apr 29 23:10:32 ah Apr 29 23:10:45 I'm making a new em-mode for the linkstations ;) Apr 29 23:10:51 neat Apr 29 23:11:13 I need to flash my HG with u-boot and put it back together Apr 29 23:11:17 so I need to fit a 2.6-kernel with xfs, ext[23] and basic utilities to support installing/fixing stuff Apr 29 23:11:30 xfs? ouch Apr 29 23:11:39 xfs *and* ext3 == huuuuge Apr 29 23:11:51 xfs is the killer though... Without xfs-support in the rootfs, I'm at 950kb. With xfs-support, I'm at 1.44mb. Apr 29 23:12:06 (This is with upx max compression, ext2 rootfs.gz) Apr 29 23:12:26 xfs is a filesystem right? what does it need in the rootfs? Apr 29 23:12:52 Utilities to create, grow, shrink, fix, diagnose etc, etc, etc. Apr 29 23:13:00 ah Apr 29 23:17:00 some of the utilities compress a lot though :) Apr 29 23:17:01 321732 -> 89928 27.95% linux/ElfPPC xfs_db Apr 29 23:17:22 on average, the xfs support utils: Apr 29 23:17:23 630820 -> 209524 33.21% [ 7 files ] Apr 29 23:17:26 heh that's pretty good Apr 29 23:20:33 * NAiL is going to do a small test Apr 29 23:20:41 I've got a freshly generated rootfs Apr 29 23:20:49 4580 root/ Apr 29 23:22:57 3380 root/ Apr 29 23:23:12 1.2mb savings Apr 29 23:23:56 how does it compare to gzip? Apr 29 23:24:31 I dunno. I haven't tried compressing the binaries with gzip. Apr 29 23:24:42 That size is the whole rootfs before compression Apr 29 23:26:04 removing an unused library, adding a missing one, then making the whole thing an ext2 fs and compressing it gives me: Apr 29 23:26:07 1508 -rw-r--r-- 1 repvik repvik 1536391 2007-04-30 01:26 rootfs.powerpc.ext2.gz Apr 29 23:26:28 That's a fully featured rootfs with ext2,ext3,xfs and dmraid support. Apr 29 23:26:37 wow, there's something I don;t understand then Apr 29 23:26:57 upx compresses the executables and then gzip does further compression? Apr 29 23:27:33 I don't think gzip can compress the upx-compressed binaries much Apr 29 23:27:45 yeah I was gonna say Apr 29 23:27:46 but gzip compresses the whole rootfs including the libraries that upx can't compress Apr 29 23:27:50 ah OK Apr 29 23:28:13 so the upx compressed binaries, they self-decompress when called? Apr 29 23:28:20 yes Apr 29 23:28:35 ok, now I begin to understand Apr 29 23:28:39 Obviously, this is a speed hit Apr 29 23:28:45 nod Apr 29 23:29:06 you save a bit of time reading them from flash, but you have to wait a bit for the decompression again Apr 29 23:29:27 the main issue is usually *fitting* everything into flash Apr 29 23:29:39 most ppl would gladly give up a bit of speed to fit more in Apr 29 23:29:41 I know I would Apr 29 23:29:55 In this case, speed really isn't an issue. Apr 29 23:30:08 It's not designed to be used except in case of emergencies. Apr 29 23:30:21 yep Apr 29 23:30:50 pretty much anything faster than mailing it off to get unbricked would be acceptable ;-) Apr 29 23:31:14 well technically not bricked Apr 29 23:31:27 hehe Apr 29 23:31:28 you'd have to remove the drive, attach to a PC and fix it Apr 29 23:31:31 yeah Apr 29 23:31:41 which is a pain, especially if you're doing development work Apr 29 23:32:11 yeah Apr 29 23:32:25 but now, with the all-singing, all-dancing u-boot and this fancy all-singing, all-dancing em-mode rootfs, this should almost be a joy ;) Apr 29 23:32:36 heh :-) Apr 29 23:33:10 btw, I'm compressing using a very old version of the upx compressor. From what I see, it's far more efficient in its later incarnations. Apr 29 23:33:54 well heck, get 2.93 then! Apr 29 23:34:32 hehe, yeah Apr 29 23:35:03 ah Apr 29 23:35:05 The main news since UPX 2 are: Apr 29 23:35:05 * new format: added support for linux elf/arm Apr 29 23:35:05 * better compression using the LZMA algorithm (option '--lzma') Apr 29 23:35:43 neat Apr 29 23:35:51 hmm, what about elf/armeb ;-) Apr 29 23:38:20 NAiL: the upx thingo couldn't get our kernel down by a full block size, so no point. Apr 29 23:40:28 I forget the block size on the nslu2 128kB, 256kB? Apr 29 23:41:06 2976 root/ Apr 29 23:41:22 down from 4850kb, that's not bad ;) Apr 29 23:41:43 lol not half bad Apr 29 23:42:38 1352 -rw-r--r-- 1 repvik repvik 1379629 2007-04-30 01:42 rootfs.powerpc.ext2.gz Apr 29 23:42:48 That's the complete compressed rootfs Apr 29 23:43:24 considering the amount of support in that rootfs, that's quite impressive. Apr 29 23:43:59 rwhitby: how much could it compress the default kernel? I'm seeing huge gains with the new (--lzma) compression. Apr 29 23:45:12 compressing it from 1.4mb down to 1mb is in my eyes nothing short of boggling. Apr 29 23:45:31 yah Apr 29 23:46:26 NAiL: it's worth looking at again. Apr 29 23:46:37 but only if it gets it down to 768KiB-16 Apr 29 23:46:49 why that small? Apr 29 23:46:55 oh, make the 1024KiB-128KiB-16 Apr 29 23:47:02 892 I guess Apr 29 23:47:22 redboot reads 1mb no matter what, doesn't it? Apr 30 00:14:55 jacques: btw, John pointed out the speed differences Apr 30 00:14:56 Yes, --lzma is really good for small space, but it does require Apr 30 00:14:56 about 10 times the CPU cycles for decompression at startup in contrast Apr 30 00:14:56 to the upx default nrv2b or nrv2e compression.  The nrv2e and nrv2b Apr 30 00:14:56 algorithms are blindingly fast at decompressing: in the range of Apr 30 00:14:58 1/3 the speed of memcpy. Apr 30 00:18:07 NAiL: redboot reads what the length header tells it. Apr 30 00:18:46 ah, yes Apr 30 00:18:55 but the length header was at 1mb? Apr 30 00:28:27 rwhitby: anyway, I say it might be worth looking into with the lzma compression ;) Apr 30 00:37:02 1524 -rw-r--r-- 1 repvik repvik 1554467 2007-04-30 02:37 rootfs.powerpc.ext2.gz Apr 30 00:37:29 That's a complete uclibc rootfs, with ext2/ext3/xfs/nfs-support, with proftpd(!) and dropbear installed. Apr 30 00:37:54 * NAiL just can't quit being amazed. Apr 30 00:38:34 oh, and it's got devicemapper-support too Apr 30 00:38:42 so dmraid should work :) **** ENDING LOGGING AT Mon Apr 30 02:59:56 2007