On 10/11/10 01:39, Nick Cairncross wrote:
Hi!
I wouldn't think you need multiple network cards to use squid, unless
your internet connection is on or above 1GB/s. If your ISP provides you
less, I would think a regular gigabit Nic would do the job.
Your Hard Drives probably wont be fast enough to cache data on multiple
Nics anyways.
We have over 1000 Clients, and the previous setup we used, we had only 1
GB network interface of our squid. It was sitting in the DMZ, and the
connections went trough it.
It was fine. Had no connection problems.
Tibby
________________________________________
Feladó: Nick Cairncross
Hi list,
I'm looking at building a couple more 3.1.8 servers on RHEL 5.5 x86. The
servers are nicely high-powered have multiple Gb NICs (4 in total). My
previous proxy server (bluecoat) had two NICs. I understand that one was
used to listen to requests and send to our upstream accelerator and one
was used if the equivalent 'send direct' was used i.e bypass the
accelerator. Can the list make any thoughts or recommendations about the
best way to utilise the NICs for best performance? Can I achieve the same
outbound as above? Should I even bother trying to do this? User base
would be about 700 users; I'm not caching. Simple ACLs but with two
authentication helpers (depending on browser).
Cheers
Nick
Thanks Tibby for your input - sounds sensible. Net connection is fast and
wide, so gb should be ok.
In that case another question for the list: Seeing as I'm not doing ANY
caching at all and just proxying traffic are there any recommendations for
squid.conf settings that might optimise my users' experiences (other than
caching..).
Why such a thing against caching?
That is the #1 speed gain (about 3-4 orders of magnitude faster to fetch
something from RAM cache than the network).
I have fast ACLs where possible in place and my squid.conf is
as below. I'm looking for any tips on maximising memory, processes etc
memory maximization is almost all about caching.
Besides the cache_mem for in-transit objects (your "cache deny all"
causes them to drop when completed even if they need fetching again
immediately).
You could also possibly check and tune the DNS ipcache/fqdncache sizes
for more entries, and bump the auth cache size up enough to hold all
your user credentials.
from within the squid.conf so that the end user has as quick an experience
as possible - Are there any other tags I should look at using? Server spec
is a single cpu Xeon X5660 @ 2.8, 6gb 1333 ram, 250 gb R1
To start with TMF (the measurement factory) are looking into a few
things right now regards to the speed of 3.1. There are likely to be
some extra speed patches in 3.1.10 next month.
The first one being the default buffer size in src/MemBuf.h (currently
2*1024, could be upped to 64*1024 for bigger network reads).
===
http_port 8080
auth_param negotiate program /usr/lib/squid/squid_kerb_auth -r
auth_param negotiate children 80
auth_param negotiate keep_alive on
auth_param ntlm program /usr/bin/ntlm_auth
--helper-protocol=squid-2.5-ntlmssp
auth_param ntlm children 60
auth_param ntlm keep_alive on
auth_param basic program /usr/bin/ntlm_auth
--helper-protocol=squid-2.5-basic
auth_param basic children 20
auth_param basic realm Squid proxy-caching web server
auth_param basic credentialsttl 2 hours
cache_peer [upstream] parent 8080 0 no-query proxy-only no-digest default
cache_mgr [blanked]
cachemgr_passwd [blanked] all
client_persistent_connections on
#server_persistent_connections on
persistent_connection_after_error on
access_log /var/log/squid/access.log squid
cache_store_log none squid
cache_log /var/log/squid/cache.log squid
###### Delay Pool Definitions #######
# Total number of delay pools
delay_pools 1
#### ACCESS CONTROL LISTS #####
## USER-AGENT (Browser-type) ACLs
acl Java_jvm browser "/etc/squid/ACL/USERAGENTS/USER-AGENTS_JAVA.txt"
acl iTunes browser "/etc/squid/ACL/USERAGENTS/USER-AGENTS_APPLE.txt"
acl MSNMessenger browser "/etc/squid/ACL/USERAGENTS/USER-AGENTS_MSN.txt"
## USER AUTHENTICATION ACLs
acl AuthenticatedUsers proxy_auth REQUIRED
## URL DESTINATION ACLs
acl URL_ALLOWDstDomains dstdomain
"/etc/squid/ACL/URL/URL_ALLOWDstDomains.txt"
## IP ACLS ##
acl CNP_SERVERIP src 172.16.10.176
acl CNP_SERVERIP src 172.16.100.50
acl CNP_CLIENTIP src "/etc/squid/ACL/IPADDRESSES/IP_CLIENTIP.txt"
## Windows Update ACLS
acl WSUS_IP src 172.16.10.127
# LAN IP ACLs
acl CNP_172SUBNETS src 172.16.0.0/16
acl CNP_SERVERSUBNETS src 172.16.10.0/24
acl CNP_SERVERSUBNETS src 172.16.100.0/24
# Blocks CONNECT method to IP addresses (Blocks Skype amongst other things)
acl StopDirectIP url_regex ^[0-9]+\.[0-9]+\.[0-9]+\.[0-9]+
dstdom_regex for a often much shorter size of bytes duplicated and sent
out for regex scanning.
# MSN Messenger Allow IP ACL
acl IP_MSNMessenger src "/etc/squid/ACL/IPADDRESSES/IP_MSNMESSENGER.txt"
# SEND DIRECT ACLs
acl SENDDIRECT_DstDomains dstdomain
"/etc/squid/ACL/SENDDIRECT/SENDDIRECT_DSTDOMAINS.txt"
acl SENDDIRECT_IPAddresses src
"/etc/squid/ACL/SENDDIRECT/SENDDIRECT_IPADDRESSES.txt"
# CONNECT Method Direct IP ACLs
acl IP_CONNECTALLOW src "/etc/squid/ACL/IPADDRESSES/IP_CONNECTALLOW.txt"
## LOCALHOST ACLs
acl localhost src 127.0.0.1
acl to_localhost dst 127.0.0.0/8
## CACHEMGR ACL
acl manager proto cache_object
## PORTS ACLs
acl SSL_ports port 443
acl Safe_ports port 80 # http
acl Safe_ports port 8080 # http
acl Safe_ports port 21 # ftp
acl Safe_ports port 443 # https
acl Safe_ports port 70 # gopher
acl Safe_ports port 70 # gopher
acl Safe_ports port 210 # wais
acl Safe_ports port 1025-65535 # unregistered ports
acl Safe_ports port 280 # http-mgmt
acl Safe_ports port 488 # gss-http
acl Safe_ports port 591 # filemaker
acl Safe_ports port 777 # multiling http
acl RTMP_ports port 1935 # RTMP
# CONNECTION METHOD ACL
acl CONNECT method CONNECT
acl POST method POST
######### ICAP SERVER #########
acl ICAP_BYPASS dstdomain "/etc/squid/ACL/ICAP/ICAP_BYPASS_URL.txt"
## ICAP Settings
icap_enable on
icap_preview_enable on
icap_persistent_connections on
icap_send_client_ip on
icap_send_client_username on
icap_client_username_header X-Authenticated-User
icap_client_username_encode on
icap_service ss reqmod_precache 0 icap://localhost:1344/ssreqmod
adaptation_service_set c1 ss
adaptation_access c1 deny ICAP_BYPASS
adaptation_access c1 deny POST
adaptation_access c1 deny CONNECT
adaptation_access c1 allow all
###### ACCESS RULES #####
## CACHEMGR ALLOW
http_access allow manager CNP_172SUBNETS
## GLOBAL DENY RULES - Do NOT include User Authentication ACLs here
http_access deny !Safe_ports
http_access deny manager !CNP_172SUBNETS
You already allowed "manager CNP_172SUBNETS" the !CNP_172SUBNETS has no
effect and can be dropped.
This bunch of denies could be sped up somewhat:
* moving the fastest to the top (CONNECT, POST)
> http_access deny CONNECT !SSL_Ports !CNP_172SUBNETS
> http_access deny POST !SSL_Ports !RTMP_ports !CNP_172SUBNETS
* re-ordering the placement of ACL tests in each of the rest.
http_access deny MSNMessenger CNP_172SUBNETS !IP_MSNMESSENGER
http_access deny StopDirectIP !IP_CONNECTALLOW
http_access deny iTunes !CNP_172SUBNETS
http_access deny Java_jvm !CNP_172SUBNETS
For example:
http_access deny CNP_172SUBNETS !IP_MSNMESSENGER MSNMessenger
instead of:
http_access deny MSNMessenger CNP_172SUBNETS !IP_MSNMESSENGER
Why? They all start with regex right now ("browser" ACL uses regex too).
That is markedly slower than IP testing.
# CACHEMGR ALLOW
http_access allow manager CNP_172SUBNETS
# POST/CONNECT Method ALLOW #
http_access allow CONNECT CNP_172SUBNETS
http_access allow POST CNP_172SUBNETS
# USER AGENTS ALLOW #
http_access allow iTunes CNP_172SUBNETS
http_access allow Java_jvm CNP_172SUBNETS
Same speed-up ordering can be done for the above agent lines as fro teh
denials earlier.
# URL and IP DESTINATIONS NOT REQUIRING AUTHENTICATION
http_access allow URL_ALLOWDstDomains
http_access allow CNP_SERVERIP
http_access allow CNP_CLIENTIP
http_access allow WSUS_IP
## USERS AUTHENTICATION ACL##
http_access allow AuthenticatedUsers
## SEND DIRECT ALLOW
always_direct allow SENDDIRECT_DstDomains
always_direct allow SENDDIRECT_IPAddresses
## LOCAL HOST ALLOW ##
http_access allow localhost
###### DELAY ACCESS RULES ######
# First delay class pool is User allocated
delay_class 1 4
# First delay class pool is limited to 4mb i.e. all users not in
exceptions: CNP_SERVERSUBNETS
delay_parameters 1 -1/-1 -1/-1 -1/-1 4000000/4000000
#First delay pool applied to Authenticated users except those in Server
Subnet
delay_access 1 allow AuthenticatedUsers !CNP_SERVERSUBNETS
Reverse the order of these too...
delay_access 1 allow !CNP_SERVERSUBNETS AuthenticatedUsers all
(tacking "all" on the end to keep Squid quiet about the auth. Delay
pools will ignore non-authenticated users, but cache.log would get a
complaint).
delay_access 1 deny all
http_access deny to_localhost
Very much too late for this one. The above rule is a security barrier to
prevent infected or broken clients looping requests back at Squid etc.
If you need it it should be up with the "deny !Safe_ports" one.
never_direct allow all
snmp_access deny CNP_172SUBNETS
Hmm, implicit allow SNMP from *everywhere* except 172.* places.
Somehow I don't think this is what you want. But I dont have enough
knowledge of your network to say what should be there instead.
Something(s) allowed the rest denied.
cache deny all
http_access deny all
Amos
--
Please be using
Current Stable Squid 2.7.STABLE9 or 3.1.9
Beta testers wanted for 3.2.0.3
