Proxy server to support a large number of simultaneous requests

[Andrey K <ankor2023@xxxxxxxxx>]

Hello,

We need to configure a dedicated proxy server to provide caching of online video broadcasts in order to reduce the load on the uplink proxy.

Hundreds of users will access the same video-chunks simultaneously.

I developed a simple configuration for the test purposes (it is shown below).

The collapsed_forwarding option is on.

I selected a couple of cacheable resources in the internet for testing:

 - small size (~400 KB): https://ia800406.us.archive.org/13/items/romeo-y-julieta-texto-completo/Romeo%20y%20Julieta%20-%20William%20Shakespeare.pdf

 - large (~8 MB): https://ia600601.us.archive.org/10/items/Linux-Journal-2015-01/Linux-Journal-2015-01.pdf

To test simultaneous connections I am forking curl using a simple script (it is also shown below).

When I run a test (500 curl threads to https://ia800406.us.archive.org/13/items/romeo-y-julieta-texto-completo/Romeo%20y%20Julieta%20-%20William%20Shakespeare.pdf) I see lots of TCP_MISS/200 with FIRSTUP_PARENT/parent_proxy records in the logs.

A simple analysis shows a low percentage of cache hits:

cat /var/log/squid.user/access.log| grep '2023-05-29 14' | grep pdf  | awk '{print $5" " $10}' | sort | uniq -c
     24 TCP_CF_MISS/200/- HIER_NONE/-
    457 TCP_MISS/200/200 FIRSTUP_PARENT/parent_proxy
     10 TCP_MISS/200/- HIER_NONE/-
      9 TCP_SWAPFAIL_MISS/200/200 FIRSTUP_PARENT/parent_proxy

So the Hit ratio is about (500-457-9)*100/500=6.8%

Almost the same situation we see when run 200 threads:

cat /var/log/squid.user/access.log| grep '2023-05-29 15:45' | grep pdf  | awk '{print $5" " $10}' | sort | uniq -c
      4 TCP_CF_MISS/200/- HIER_NONE/-
    140 TCP_MISS/200/200 FIRSTUP_PARENT/parent_proxy
     40 TCP_MISS/200/- HIER_NONE/-
     16 TCP_SWAPFAIL_MISS/200/200 FIRSTUP_PARENT/parent_proxy

This time the Hit ratio is about (200-140-16)*100/500=21%

With 50 threads the Hit ratio is 90%:

cat /var/log/squid.user/access.log| grep '2023-05-29 15:50' | grep pdf  | awk '{print $5" " $10}' | sort | uniq -c
     27 TCP_CF_MISS/200/- HIER_NONE/-
      1 TCP_MISS/200/200 FIRSTUP_PARENT/parent_proxy
     18 TCP_MISS/200/- HIER_NONE/-
      4 TCP_SWAPFAIL_MISS/200/200 FIRSTUP_PARENT/parent_proxy

I thought that it should always be near 99% - only the first request to an URL should be forwarded to the parent proxy and all subsequent requests should be served from the cache.

The situation is even worse with downloading a large file:

500 threads (0.4%):

cat /var/log/squid.user/access.log| grep '2023-05-29 17:2' | grep pdf  | awk '{print $5" " $10}' | sort | uniq -c
     10 TCP_CF_MISS/200/200 FIRSTUP_PARENT/parent_proxy
      2 TCP_CF_MISS/200/- HIER_NONE/-
    488 TCP_MISS/200/200 FIRSTUP_PARENT/parent_proxy

200 threads (3%):

cat /var/log/squid.user/access.log| grep '2023-05-29 17:3' | grep pdf  | awk '{print $5" " $10}' | sort | uniq -c
      9 TCP_CF_MISS/200/200 FIRSTUP_PARENT/parent_proxy
      6 TCP_CF_MISS/200/- HIER_NONE/-
    180 TCP_MISS/200/200 FIRSTUP_PARENT/parent_proxy
      5 TCP_SWAPFAIL_MISS/200/200 FIRSTUP_PARENT/parent_proxy

50 threads (98%):

 cat /var/log/squid.user/access.log| grep '2023-05-29 17:36' | grep pdf  | awk '{print $5" " $10}' | sort | uniq -c
     25 TCP_CF_HIT/200/- HIER_NONE/-
     12 TCP_CF_MISS/200/- HIER_NONE/-
     12 TCP_HIT/200/- HIER_NONE/-
      1 TCP_MISS/200/200 FIRSTUP_PARENT/parent_proxy

Could you clarify if this behavior of my squid is a bug/misconfiguration, or if I'm running into server performance limitations (squid is running on a VM with 22 cores)?

Kind regards,

     Ankor

squid.conf:

workers 21

sslcrtd_program /data/squid.user/usr/lib/squid/security_file_certgen -s /data/squid.user/var/lib/squid/ssl_db -M 20MB
sslcrtd_children 21

logformat extended-squid %{%Y-%m-%d %H:%M:%S}tl| %6tr %>a %Ss/%03>Hs/%<Hs %<st %rm %ru %un %Sh/%<A %mt %ea

logfile_rotate 0
access_log daemon:/var/log/squid.user/access.log logformat=extended-squid on-error=drop

cache_peer parent_proxy  parent 3128 0
never_direct allow all

cachemgr_passwd pass config

acl PURGE method PURGE
http_access allow PURGE

http_access allow all

http_port 3131 ssl-bump generate-host-certificates=on dynamic_cert_mem_cache_size=20MB tls-cert=/etc/squid.user/sslbump/bump.crt tls-key=/etc/squid.user/sslbump/bump.key
sslproxy_cert_error allow all

acl step1 at_step SslBump1
acl step2 at_step SslBump2
acl step3 at_step SslBump3

ssl_bump peek step1
ssl_bump bump step2
ssl_bump bump step3

cache_dir rock /data/squid.user/cache 20000 max-size=12000000
cache_swap_low 85
cache_swap_high 90

collapsed_forwarding on

pinger_enable off
max_filedesc 8192
shutdown_lifetime 5 seconds
netdb_filename none
log_icp_queries off
client_request_buffer_max_size 100 MB

via off
forwarded_for delete

coredump_dir /data/squid.user/var/cache/squid

curl_forker.sh:

#!/bin/sh
N=100
URL="" href="https://ia600601.us.archive.org/10/items/Linux-Journal-2015-01/Linux-Journal-2015-01.pdf">https://ia600601.us.archive.org/10/items/Linux-Journal-2015-01/Linux-Journal-2015-01.pdf

if [[  -n $1 &&  $1 =~ help$  ]];
then
   echo "Usage: $0 [<cnt>] [<url>]"
   echo
   echo "Example: $0 10 https://ia600601.us.archive.org/10/items/Linux-Journal-2015-01/Linux-Journal-2015-01.pdf";
   echo
   exit;
fi

while [[ $# -gt 0 ]]
do
  if [[ $1 =~ ^[0-9]+$ ]]
  then
     N=$1
  else
     URL="">  fi
  shift
done

echo $URL
echo $N threads

for i in `seq $N`
do
  nohup curl --tlsv1.2 -k   --proxy 0001vsg01:3131  -v $URL  >/dev/null  2>&1 &
done

_______________________________________________
squid-users mailing list
squid-users@xxxxxxxxxxxxxxxxxxxxx
http://lists.squid-cache.org/listinfo/squid-users