about:config?network.http.pipelining.maxrequests=1 000

Just making sure, is this 1000 or 1 000?

PS: with the 1st 4 about:config settings browser performance was improved visibly. -- Thanks

I did the MicroB tweaks. I must say that the browser seems to load pages faster.

I'd like to try the kernel tweaks, but I'm not really a Linux guy.

Hi Brontide and everyone else.
I have seen 3 different versions of brontide's sysctl changes (the one at the start of this thread, the large one in page 2, and the one in the deb package) each different in a seemingly random fashion, so I wanted to have a deeper look at those sysctl options and their meaning, and compare the changes with the defaults.

The sysctls are documented here:
http://ipsysctl-tutorial.frozentux.n...tml/index.html
http://www.mjmwired.net/kernel/Docum.../sysctl/vm.txt
http://www.mjmwired.net/kernel/Docum...stems/proc.txt

Follows my discussion (the sysctls show their default value in the tablet, followed by the 3 brontide values):

net.ipv4.tcp_window_scaling = 1 (0 1 0) #Measured transferring a 50 MB file from server to tablet (to /dev/null) using scp (I got better throughput and less variance than using plain ftp) over sort and long distance (about 25 m, not straigth) on a 54Mbps wifi. Couldn't measure any noticeable difference. It's value shouldn't matter, since this only affects connections with high throughput and high roundtrip time, that is, the tcp window is only scaled when it is not big enough. I think the default (enabled) is the best.

#Selective ACKs are an advanced tcp capability, really good to use when packets are easily lost (like wifi), they are fully backwards compatible and shouldn't be changed. Let them at the default (enabled).
net.ipv4.tcp_sack = 1 (0 1 1)
net.ipv4.tcp_dsack = 1 (0 1 1)
net.ipv4.tcp_fack = 1 (- - 1)

net.ipv4.tcp_keepalive_time = 7200 (1800 1800 1800) #this is a ping sent every so many second (2 hours) to keep a connection going when it has no traffic. It doesn't change anything if you lower this value to 30 min. Doesn't matter at all.

net.ipv4.tcp_fin_timeout = 60 (10 10 10) #this is the time in seconds to drop closed connections if the peer doesn't respond. Each connection uses non-swapable memory, it is a good idea to lower this value if you use any p2p program. If browsing it doesn't matter that much. 10 is good, I'd test even lower values if using p2p.

net.ipv4.tcp_timestamps = 1 (0 0 0) # good on links >10Mbps, bad only on very low speed links (like 54 kbps modems). I'd left it turned on (default).

net.ipv4.tcp_max_syn_backlog = 128 (- 96 -) #this is how many outgoing connections you can start at once. Changing this value not much will not make any difference, lowering it too much will cause problems.

net.ipv4.tcp_ecn = 0 (- 1 -) #congestion notification, useful for routers, I think it doesn't matter in the tablet at all.

#network buffers. Default values are pretty curious, since max tcp values are bigger than core values and core values override the max tcp values, but the same happens on my opensuse. Core max values are bit lower than the usual default values, probably because Nokia developers lowered them to fit better in the tablet. Increasing them doesn't have any effect in throughput and can lead to worse memory usage.
net.core.rmem_max = 105472 (- 2097152 -)
net.core.wmem_max = 105472 (- 2097152 -)
net.ipv4.tcp_rmem = 4096 87380 258048 (- [16384 65536 2097152] -)
net.ipv4.tcp_wmem = 4096 16384 258048 (- [16384 32768 2097152] -)

net.ipv4.tcp_reordering = 3 (- - 5) # How much a packet may come out of order in the tcp stream before assuming it has been lost. Packets may come out of order only in the long and complex pipes of internet, not from the wifi AP. Increasing it won't have any possitive effect in the tablet. Left it at the default.

net.ipv4.tcp_synack_retries = 5 (- - 3) # Retries to start a new connection (5 takes about 180 seconds), lowering this a bit will take less time to give an error when the connection can't be initiated. It always annoyed me how long it takes to give the error message when the connection doesn't work, so I think lowering this to 3 (100 seconds) is a good idea (makes the network experience more interactive)

#Those are good at the default, no point in playing with them.
net.ipv4.icmp_echo_ignore_broadcasts = 1 (- - 1)
net.ipv4.icmp_ignore_bogus_error_responses = 1 (- - 1)

vm.swappiness = 1 (25 50 100) # I agree that increasing a bit the swap usage may be a good idea to increase responsiveness, but it will increase flash wearing so I think it shouldn't be used without first setting a swap file in the external flash card. My opensuse server has it at 60, so I think reasonable values for the tablet should be lower than that.

vm.vfs_cache_pressure = 100 (150 200 10000) #increasing this will reduce the caching of inodes. I think it will provide nor benefit nor perjudice to change this value (there are no seek penalty for disk access). Extensive testing would be needed to see the real effect.

vm.min_free_kbytes = 1442 (768 512 4096) #How much memory must be kept free for page fault memory allocations. Never lower than 1024, never too big. My opensuse server has 3816, so may be the default is a good value.

# Those affect swapping, may be good to play with.
vm.dirty_background_ratio = 60 (- 5 -) (opensuse = 5)
vm.dirty_ratio = 120 (- 20 -) (opensuse = 10)
vm.page-cluster = 3 (- 2 -) (opensuse = 3)

Summarizing:
net.ipv4.tcp_fin_timeout = 10 (even lower if using p2p)
net.ipv4.tcp_synack_retries = 3
vm.vfs_cache_pressure = x (test what happens with x>100)
vm.swappiness = x (x << 60)
vm.dirty_background_ratio = x (5<=x<=60)
vm.dirty_ratio = x (10<=x<=120)
vm.page-cluster = x (0<=x<=3)
swap file should be on external mmc if swapping is increased.

thanks for the info, most of my work has benn trial and error based on what I have learned from laptops over the years

these are great stuff that you folks are sharing!

just wondering if these still work after the current firmware updates?

I miss brontide.

+1

... his threads always gave me something to do that was a learning experience... Sort of like Bill Nye the Science Guy, LoL.

I know this is an older post but has anyone managed to make an easy deb file for these browser tweaks? Ive read this thread 10 times now and can say without a doubt i have NO CLUE how to apply these settings! Is there any easier way to get a faster browser? Id love to speed up this system any way possible!

Any help would be appreciated

Adblock will noticeably slow down the browser. If you have that addon uninstall and replace with one of the host file solutions available.

The way to speed up your browser is to use tear instead. Tear and, as Fragos said, a host file.