Hi,
If you've got over 650 EAS users, a PREFORK of 400 will never work.
EAS clients will fight to get sogod workers and will leave nothing to
the SOGo web interface.
Dedicate a SOGo instance to EAS and use a handful of workers for the web
interface. See
https://sogo.nu/support/faq/dedicated-separate-sogo-instance-for-activesync.html
for details.
Ludovic
On 2019-09-04 11:30 a.m., bdus...@luzerne.edu wrote:
A few months ago we began providing SOGo for our students. It was
summer and the number of students using the server was limited. I
found tuning was needed, especially the prefork value. I was able to
find good values for our installation within a day or two.
We started a new semester a few days ago and our full student body is
using the product. We've encountered performance problems several
times a day. I've handled these problems by making a tuning change
and rebooting the server. Within a few hours the problem returns.
I haven't been able to find appropriate configuration settings for our
environment.
We have a little over 14,000 users. I wrote some code which
attempts to determine how many users are using ActiveSync. Searching
the nginx access logs I'm finding about 650 unique authenticated
accounts posting to /Microsoft-Server-ActiveSync?User=. I'm only
searching logs from the past seven days. I've also attempted to get
a rough estimate of current web users by using:
netstat | grep http | wc -l
I get about 700 returned. I know this isn't completely accurate, but
it gives me an estimate.
The server running SOGo is Centos 7.6. It's virtualized with 8 CPU
and 16 GB of ram. We're a VMware environment. Checking the server
performance doesn't show anything significant. CPU is usually around
10% with some spikes. RAM usage is <10GB.
Currently my SOGO sysconfig file contains:
PREFORK = 400
SOGoMaximumPingInterval = 3540
SOGoMaximumSyncInterval = 3540
SOGoInternalSyncInterval = 30
WOWatchDogRequestTimeout = 5
WOListenQueueSize = 50
I've modified the OS kernel parameters as follows in the sysctl.conf file:
net.ipv4.tcp_max_syn_backlog=8192
net.ipv4.tcp_fin_timeout=25
net.core.somaxconn=1280
I run NetData on this server. It was indicating problem with tcp
accept queue overflows and drops. This is why I made the changes to
sysctl.conf.
It's now indicating tcp syn cookies queue is overflowing. I'm going
to do some research on this and will make appropriate changes.
Does anyone have any suggestions? I'm making changes, but the
performance doesn't appear to be improving much.
Thanks,
Bob
Bob Dushok
Director of Enterprise Systems and Computer Labs
Luzerne County Community College
1-800-377-5222 ext 7327
bdus...@luzerne.edu
--
users@sogo.nu
https://inverse.ca/sogo/lists
--
Ludovic Marcotte
lmarco...@inverse.ca :: +1.514.755.3630 :: https://inverse.ca
Inverse inc. :: Leaders behind SOGo (https://sogo.nu), PacketFence
(https://packetfence.org) and Fingerbank (https://fingerbank.org)
--
users@sogo.nu
https://inverse.ca/sogo/lists
