Galaxis
Member
- Joined
- Aug 30, 2010
- Messages
- 318
Yesterday, I was at a conference, and used my Pandora a whole day for the first time (uptime in the end was about 9 hours, with maybe 6 hours out of standby, and a nick below 20% battery left). Now I'm probably not a typical user, because I was just using two applications - a Terminal (with several ssh sessions), and Firefox for web browsing on the conference wireless, with the system booting off an SD card.
Terminal and ssh are unintrusive, and I could probably use as many of them as I wanted (although Terminal is not exactly lightweight for what it does). Firefox on the other hand - well, it's still my favourite browser, but it's just not made for devices with relatively little memory. Maybe things get better with the memory optimizations in FF7.
For example, when accessing Google+ (couldn't find a way to the mobile version with a fully fledged browser), it absolutely needs swap, and you can practically see the system burning through battery. At one point, I had to wait for about 30 seconds until Linux found out of it's memory starvation - yes, swapping to SD cards is slow - but it didn't crash, and afterwards, things were usable until the end of the day.
Some statistics I did get in the afternoon (not as useful as I thought they might be, but hey):
Disk I/O stats. mmcblk0p1 is a small FAT partition for autoboot.txt, 0p2 is an ext3 partition with the system (and PND directories), 0p3 is swap space.
It's usually said that swap gets lots more reads than writes - there's some indication of that (7441 reads vs. 5274 writes - and the number of sectors read is actually lower than those written), but not really as big a difference as I thought. Seems that most of the stuff the kernel has thrown out of RAM actually isn't used all that much (and so swapping is particulary effective, as memory is freed for data that's in use). Also note that the system partition has more than three times the writes of swap, which I assume is mostly FFs disk cache. So it's definitely not the swapping that's going to kill my SD card
(On diskstats fields, see here - although the kernel documentation doesn't seem 100% correct, partition fields are the same as for the device.)
Snippet of top output - about 110M of swap used:
In the end - yay: No more lugging the notebook around. Sad: Several people asked about the Pandora and I could only tell them they may register for a preorder, to possibly get one in january.
Terminal and ssh are unintrusive, and I could probably use as many of them as I wanted (although Terminal is not exactly lightweight for what it does). Firefox on the other hand - well, it's still my favourite browser, but it's just not made for devices with relatively little memory. Maybe things get better with the memory optimizations in FF7.
For example, when accessing Google+ (couldn't find a way to the mobile version with a fully fledged browser), it absolutely needs swap, and you can practically see the system burning through battery. At one point, I had to wait for about 30 seconds until Linux found out of it's memory starvation - yes, swapping to SD cards is slow - but it didn't crash, and afterwards, things were usable until the end of the day.
Some statistics I did get in the afternoon (not as useful as I thought they might be, but hey):
Disk I/O stats. mmcblk0p1 is a small FAT partition for autoboot.txt, 0p2 is an ext3 partition with the system (and PND directories), 0p3 is swap space.
It's usually said that swap gets lots more reads than writes - there's some indication of that (7441 reads vs. 5274 writes - and the number of sectors read is actually lower than those written), but not really as big a difference as I thought. Seems that most of the stuff the kernel has thrown out of RAM actually isn't used all that much (and so swapping is particulary effective, as memory is freed for data that's in use). Also note that the system partition has more than three times the writes of swap, which I assume is mostly FFs disk cache. So it's definitely not the swapping that's going to kill my SD card
Code:
# fgrep mmcblk0 /proc/diskstats
179 0 mmcblk0 27449 28314 1025650 710343 24970 156924 1450136 99032289 87 2397937 100090695
179 1 mmcblk0p1 109 495 2056 1164 0 0 0 0 0 890 1164
179 2 mmcblk0p2 19897 4924 780882 435351 19696 127679 1173840 36291304 87 2345859 37039046
179 3 mmcblk0p3 7441 22880 242576 273453 5274 29245 276296 62740984 0 670875 63050109
(On diskstats fields, see here - although the kernel documentation doesn't seem 100% correct, partition fields are the same as for the device.)
Snippet of top output - about 110M of swap used:
Code:
top - 16:13:20 up 6:38, 4 users, load average: 0.18, 2.09, 2.04
Tasks: 96 total, 1 running, 95 sleeping, 0 stopped, 0 zombie
Cpu(s): 11.2%us, 5.2%sy, 0.0%ni, 78.7%id, 4.7%wa, 0.2%hi, 0.1%si, 0.0%st
Mem: 248844k total, 222540k used, 26304k free, 1432k buffers
Swap: 500248k total, 114208k used, 386040k free, 62460k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
1130 root 20 0 31660 14m 2804 S 9.5 5.9 4:41.45 X
1300 user 20 0 46588 7704 5664 S 1.9 3.1 0:31.62 Terminal
2209 user 20 0 451m 130m 30m S 0.0 53.8 44:40.64 firefox-bin
2535 user 20 0 82736 13m 10m S 0.0 5.6 1:45.53 plugin-containe
In the end - yay: No more lugging the notebook around. Sad: Several people asked about the Pandora and I could only tell them they may register for a preorder, to possibly get one in january.
Last edited by a moderator: