Caching proxy w/ polipo (part1)

Let's see if we can (at least make it seem like) web browsing is faster...
(i'm still not sure whether installing SQM was an improvement)

I chose polipo as it is geared to home use and is supposedly easier to deal with than squid.
And it comes with its own luci app so install is dead easy
install luci-app-polipo
defaults looked sensible, so I just directed my browser to use router.lan:8123 as proxy, and presto - things indeed 'feel' faster. A check with chrome didn't reveal a significant decrease in load time though - 11.9s vs 11, even on a reload... but in this case the log showed a lot of cache directives not being recognized by polipo.
! there might be an issue with the resolver and local hostnames (openwrt.lan) but not sure yet.

Anyhoo, so let's add some storage for polipo to use so everything won't have to go in RAM.
Apparently ext4 is the better filesystem for wrt so let's prepare a usb stick for that.
(sudo)
fdisk -l -> check where your stick is at e.g. /dev/sdb1
fdisk /dev/sdb1
d (delete) n (new) p (primary) 1 (first partition) - accept the defaults - w (write)
mkfs.ext4 /dev/sdb1
some black magic that is supposed to make things faster by switching off journaling (wich should make the first line redundant IMO)(something fishy with -o vs -O as well)
tune2fs -o +journal_data_writeback /dev/sdb1
tune2fs -O +has_journal /dev/sdb1
tune2fs -m 0 /dev/sdx1

Last line doesn't reserve space for priviliged processes.
Then just mount the stick via luci system>mount points>add
and point polipo to the same location service>polipo>logging and ram>on-disk cache.
Apparently there's no limit setting for the size used, only how quickly files should be deleted. let's keep our fingers crossed. :)

next, we should point all our http traffic to our proxy.
hardcore way is to use iptables to redirect anything from lan with dport 80 to port 8123. But that makes it impossible for a user to decide whether to use the proxy or not - not so smart when we're still just trying things out.... part2 soon on dhcp/wpad/pac file.



Comments

Post a Comment

Popular posts from this blog

Traffic accounting

Got this from here: http://www.catonmat.net/blog/traffic-accounting-with-iptables/ All i'm interested in is my monthly bandwidth consumption, not the individual devices, there are plenty of forum threads/tools to do that. But why bring in additional tools when iptables has it all? iptables -N TRAFFIC_ACCT iptables -I FORWARD -j TRAFFIC_ACCT iptables -A TRAFFIC_ACCT and to just get the number: iptables -L TRAFFIC_ACCT -v | sed -n 3p | cut -d' ' -f2 to reset the counters /usr/sbin/iptables -L TRAFFIC_ACCT -Z the cron job to reset at the begining of the month, and a daily telegram msg 1 0 1 * * /usr/sbin/iptables -L TRAFFIC_ACCT -Z  0 8 * * * /usr/bin/curl -s -X POST https://api.telegram.org/<BOTID>/sendMessage -d chat_id=<CHATID> -d text="$(/usr/sbin/iptables -L TRAFFIC_ACCT -v | /bin/sed -n 3p | /usr/bin/cut -d' ' -f3)" >/dev/null 2>&1 (you'll need to opkg install ca-bundle for the https to work)
Read more

QoS with SQM (and something about bufferbloat)

another feature that came with stock zyxel fw is their 'optimized' (the one with the 400mbps nag) QoS packages. As far as i can tell the standard SQM package should come pretty close. install uci-app-sqm a Network>SQM QoS menu item will appear check enable, double check your interface name is really your wan port(eth1), and set your down/up speeds to what you measured with e.g. dslreports . under linklayer adaptation set to Ethernet (for VDSL) with 8 bytes overhead. i reran the test, and it seems my up/down throughput dropped about 10%,  but my bufferbloat report went from D to B.... whatever that means in reality I'll have to find out :)
Read more

Ath10k alternative driver firmware - CC15.05.1 wireless unstable

I'm still experiencing trouble with the 5Ghz, maybe even WRT in general, up to the point I'm considering going back to the original Zyzel AA release. One last effort to see if this helps: I'm currently running an alternative driver firmware for the  QCA988X get it from github i'm running the last of the 4 series 10.2.4.48 (not sure if running the 5 series is possible, something for later) rename the file to firmware-4.bin and replace the one that's already there in /lib/firmware/ath10k/QCA988X/hw2.0 There's probably a way to force WRT to reload the driver, but hey I rebooted. First look seems fine, let's see if it's more stable. -- update -- leaving channel selection on 'auto' didn't work - 5G did not come up. (no mesgs as to why) manually selecting a channel did seem to work, at least when also selecting a 20Mhz bandwidth, 40Mhz seemed to work but I had some devices not connecting properly. -- update2 -- 5G was mostly s...
Read more