Improve Custom refresh pattern

aGeekhere

You can use squidlite to check if the refresh pattens are improving the hit rate. I have found that the comodo antivirus updates were caching ( cannot retest at the moment) and so did window updates. The best thing is to test if the hit ratio is improving.

saluto

How did he do that?
https://www.youtube.com/watch?v=18QSf6jb1eM

Valeriy

@KOM

Here are changes made to squid.conf (or through GUI on pfsense in corresponding custom fields)
that allows caching of Windows Update for me:


acl Windows_Update dstdomain windowsupdate.microsoft.com
acl Windows_Update dstdomain .update.microsoft.com
acl Windows_Update dstdomain download.windowsupdate.com
acl Windows_Update dstdomain www.download.windowsupdate.com
acl Windows_Update dstdomain au.download.windowsupdate.com
acl Windows_Update dstdomain bg.v4.pr.dl.ws.microsoft.com
range_offset_limit 200 MB Windows_Update
quick_abort_min 1000 KB
quick_abort_max 5000 KB
quick_abort_pct 80
request_body_max_size 0 KB

refresh_pattern -i .*windowsupdate.com/.*\.(cab|exe)                     259200 100% 259200 ignore-no-store ignore-reload reload-into-ims
refresh_pattern -i .*update.microsoft.com/.*\.(cab|exe|dll|msi|psf)                  259200 100% 259200 ignore-no-store ignore-reload reload-into-ims
refresh_pattern -i windows.com/.*\.(cab|exe|ms[i|u|f]|[ap]sf|wm[v|a]|dat|zip) 4320 80% 43200 reload-into-ims
refresh_pattern windowsupdate.com/.*\.(cab|exe|dll|msi|psf) 10080 100% 43200 reload-into-ims
refresh_pattern download.microsoft.com/.*\.(cab|exe|dll|msi|psf) 10080 100% 43200 reload-into-ims
refresh_pattern www.microsoft.com/.*\.(cab|exe|dll|msi|psf) 10080 100% 43200 reload-into-ims
refresh_pattern au.download.windowsupdate.com/.*\.(cab|exe|dll|msi|psf) 4320 100% 43200 reload-into-ims
refresh_pattern bg.v4.pr.dl.ws.microsoft.com/.*\.(cab|exe|dll|msi|psf) 4320 100% 43200 reload-into-ims

cache_mem 15000 MB 
# Should be less than 50% of avail. RAM

maximum_object_size_in_memory 5000000 KB 
#should be less than cache_mem 

memory_replacement_policy heap LFUDA
cache_replacement_policy heap LFUDA

minimum_object_size 1000 KB
#Do not store very small object on disk

maximum_object_size 5000 MB 
#should be less or equal to maximum_object_size_in_memory

Here is output for today access.log of squid (number of HITS):
cat /var/squid/logs/access.log | grep HIT | grep windowsupdate -c
2860
cat /var/squid/logs/access.log | grep HIT | grep microsoft.com -c
19128

Here is output for today access.log of squid (number of MISSes):
cat /var/squid/logs/access.log | grep MISS | grep windowsupdate -c
61834
cat /var/squid/logs/access.log | grep MISS | grep microsoft.com -c
55567

I had to rebuild my disk cache 2 days ago, so I assume not everything is cached yet. Or some of my settings needs improvement…

It might not be ideal setup.
If you have some comments, improvements, please let me know.

Regards,

Valeriy

saluto

How did your dynamic cache get now?

Valeriy

@saluto:

How did your dynamic cache get now?

It is good.
This is result for yesterday:


[2.3.3-DEVELOPMENT]/var/squid/logs: cat /var/squid/logs/access.log.2 | grep MISS | grep microsoft.com -c
96270
[2.3.3-DEVELOPMENT]/var/squid/logs: cat /var/squid/logs/access.log.2 | grep HIT | grep microsoft.com -c
21177
(18% HITS vs. 82% MISS)
[2.3.3-DEVELOPMENT]/var/squid/logs: cat /var/squid/logs/access.log.2 | grep HIT | grep windowsupdate -c
8352
[2.3.3-DEVELOPMENT]/var/squid/logs: cat /var/squid/logs/access.log.2 | grep MISS | grep windowsupdate -c
34794
(19% HITS vs. 81% MISS)

KOM

:o

81% miss rate seems awfully high for something that's supposedly working good.

aGeekhere

Should look at the hit/miss ratio of file sizes. If you miss 99% of 1kB files it does not matter however it can through your ratio out.

saluto

Has any colleague been able to implement this dynamic cache, Squid-3.5 in PFSense?
https://github.com/hscbrasil/hsc-dynamic-cache
http://comastnet.blogspot.com.br/2016/06/squid-youtube-cache-2016-terbaru-suport.html

SaschaITM

KOM: "19% hit rate seems awfully high for something that's supposedly not working at all." (;

Even 19 % hit rate is a benefit if you have to work with a slow/congested line. I think hit rate can be optimized by setting up a machine that will download updates before the standard time frame windows is using to check for/download updates, so that updates are already cached when other machines check for them (I'm going to implement this on one of the systems I'm managing as soon as I find the time). Also, like aGeekHere said, you'd have to look at the byte ratio to be able to see how effective this form of caching actually is. Not sure how this could be done, maybe with a tool like Calamaris and a pre-filtered Squid log file?

Bottom line: caching Windows updates does indeed work, and is actually useful for (at least) some people.

KOM

Bottom line: caching Windows updates does indeed work

I still have yet to see any evidence to back this claim. A 19% hit rate for WU is crap. And no, there aren't a zillion little 1K files skewing the ratio.

Here is a proper test:

take two Windows 7 clients that you installed fresh and put them both behind the proxy, leaving only 1 powered on.
Install all updates until there are none left
Power on the second unit and then do all the same updates
Once complete, on pfSense run:

squidclient -h LAN_IP -p 3128 mgr:info

and post the output. That will give a lot more detail about the recent session.

aGeekhere

updated, thanks to Valeriy

if there are any mistakes in the refresh pattern let me know

Valeriy

KOM was right, in fact it was not working properly. Also, the results of bare counting of MISSes/HITs from not fully populated cache was not good measurement. Right now, any windowsupdate downloaded file (even partial) is fetched from cache, verified manually and monitored in access.log.

Few days ago I have made some improvements to utilize store_rewrite module.
They were tested, as suggested above by KOM, successfully.

Indeed, some of WU were not cached properly since they were received from different CDN hosts.

Config files are below (squid.conf is only part that is related to this issue)

Apple and Symantec updates are cached well, too.

Steam/Xbox (cs.steampowered.com / llnwd.net respective CDN) are not confirmed yet.

The rest of refresh patterns and URL rewrites are still in experimental mode. Needs a lot of testing and research in order to make it done properly.

Attached is screenshot from Lightsquid, you can clearly see changes after 3rd of March.

P.S. Questions:

not quite sure if I need to list all possible domains under acl dstdomain directive, or just first level domain will be good enough?

Comments:

I have total 1.3TB of swap available and 32GB on server running Pfsense+Squid+SquidGuard+Snort+PfBlocker. Configuration is optimized for ~2000 hosts on LAN.

Current byte/hit rate is 20-45%

/usr/local/etc/squid/squid.conf


.......
# No range offset limit for windowsupdate sites
acl Windows_Update dstdomain windowsupdate.microsoft.com
acl Windows_Update dstdomain update.microsoft.com
acl Windows_Update dstdomain download.windowsupdate.com
acl Windows_Update dstdomain windowsupdate.com
acl Windows_Update dstdomain au.download.windowsupdate.com
acl Windows_Update dstdomain bg.v4.pr.dl.ws.microsoft.com
acl Windows_Update dstdomain wsus.ds.download.windowsupdate.com
acl Windows_Update dstdomain au.b1.download.windowsupdate.com
range_offset_limit -1  Windows_Update

# Storeid_rewrite configuration

acl updatesites dstdom_regex "/usr/local/etc/squid/updatesites.txt"

store_id_access allow updatesites
store_id_access deny all
store_id_program /usr/local/libexec/squid/storeid_file_rewrite /usr/local/etc/squid/storeid_rewrite.conf
store_id_children 200 startup=60 idle=1 concurrency=0

# Refresh patterns
# 525600 min is one year

# Gaming CDN

refresh_pattern -i \.llnwd.net 525600 100% 525600 reload-into-ims ignore-reload override-expire override-lastmod ignore-no-store ignore-private ignore-auth

refresh_pattern -i \.cs.steampowered.com 525600 100% 525600 reload-into-ims ignore-reload override-expire override-lastmod ignore-no-store ignore-private ignore-auth

#windows update

refresh_pattern -i \.update.microsoft.com/.*\.(cab|exe|ms[i|u|f]|[ap]sf|wm[v|a]|dat|zip) 525600 100% 525600 reload-into-ims ignore-reload override-expire override-lastmod ignore-no-store ignore-private ignore-auth

refresh_pattern -i \.windowsupdate.com/.*\.(cab|exe|ms[i|u|f]|[ap]sf|wm[v|a]|dat|zip) 525600 100% 525600 reload-into-ims ignore-reload override-expire override-lastmod ignore-no-store ignore-private ignore-auth

refresh_pattern -i \.download.microsoft.com/.*\.(cab|exe|ms[i|u|f]|[ap]sf|wm[v|a]|dat|zip) 525600 100% 525600 reload-into-ims ignore-reload override-expire override-lastmod ignore-no-store ignore-private ignore-auth

refresh_pattern -i \.ws.microsoft.com/.*\.(cab|exe|ms[i|u|f]|[ap]sf|wm[v|a]|dat|zip) 525600 100% 525600 reload-into-ims ignore-reload override-expire override-lastmod ignore-no-store ignore-private ignore-auth

# Symantec Updates
# Only reload-into-ims works properly, other flags break update process 

refresh_pattern -i \.symantecliveupdate.com/.*\.(zip|exe|z7)          525600 100% 525600 reload-into-ims

/usr/local/etc/squid/updatesites.txt


\.adobe.com
\.java.com
\.sun.com
\.oracle.com
\.apple.com
\.microsoft.com
\.windowsupdate.com
\.ubuntu.com
\.steampowered.com
\.llnwd.net
\.symantecliveupdate.com

/usr/local/etc/squid/storeid_rewrite.conf


^http:\/\/.+?\.microsoft\.com\/.+?_([0-9a-z]{40})\.(cab|exe|ms[i|u|f]|asf|wm[v|a]|dat|zip|psf|appx)     http://wupdate.squid.local/$1
^http:\/\/.+?\.windowsupdate\.com\/.+?_([0-9a-z]{40})\.(cab|exe|ms[i|u|f]|asf|wm[v|a]|dat|zip|psf|appx) http://wupdate.squid.local/$1
^http:\/\/.+?\.cs\.steampowered\.com\/(.*)      http://steamupdates.squid.internal/$1
^http:\/\/.+?\.apple\.com\/(.*) http://appupdates.apple.squid.internal/$1
^http:\/\/.+?\.llnwd\.net\/(.*)                 http://llnwd.net.squid.internal/$1

Regards,

Valeriy

Valeriy

Just a note: manually modifying squid.conf is not good idea, since it can and will be rewritten by PfSense later. With next post I ll try to summarize all the changes and how to incorporate them using GUI of pfSense.

SaschaITM

This is interesting, I wish I had the time to test this thoroughly. Will do that as soon as I find the time.

FWIW, I had massive problems with the "range_offset_limit -1" directive and Windows 10 updates, with "psf" added to the "refresh_pattern". Update download was extremely slow and used up all available bandwidth after a while. It looked like Win10 was aborting and restarting the update download every few seconds, while Squid was still downloading from previous requests. Commented out the "range_offset_limit -1", and updates were downloading "normally" again. The "-1" value isn't documented in the current Squid docs, but it's documented for Squid 2.7. I wonder if "range_offset_limit -1" is still valid/needed with recent Squid versions.

saluto

It worked very well.

saluto

No caches of video, can you help?

Valeriy

Forget about video caching…
Most video services push content via HTTPS, which we cannt cache.

If you really need to have some video supplied to your users, then just download it and make is available via PLEX service, for instance or just regular network file system (SMB, NFS, etc.)

pfsensation

@Valeriy:

Forget about video caching…
Most video services push content via HTTPS, which we cannt cache.

If you really need to have some video supplied to your users, then just download it and make is available via PLEX service, for instance or just regular network file system (SMB, NFS, etc.)

Unless…You do MITM inspection.

saluto

@pfsensation:

@Valeriy:

Forget about video caching…
Most video services push content via HTTPS, which we cannt cache.

If you really need to have some video supplied to your users, then just download it and make is available via PLEX service, for instance or just regular network file system (SMB, NFS, etc.)

Unless…You do MITM inspection.

I do not know the MITM inspection.

pfsensation

@saluto:

@pfsensation:

@Valeriy:

Forget about video caching…
Most video services push content via HTTPS, which we cannt cache.

If you really need to have some video supplied to your users, then just download it and make is available via PLEX service, for instance or just regular network file system (SMB, NFS, etc.)

Unless…You do MITM inspection.

I do not know the MITM inspection.

I'm currently having issues with it after the latest Squid update, unsure as to why. However, you can do MITM inspection by enabling SSL filtering and deploying out a CA certificate to your clients.