High Server Load¶
If your server load is high (server loads that are <= the number of CPU Cores in your server are quite normal), you need to locate the bottleneck and make optimizations targeting it.
Most of the time, high server load is caused by PHP processes and MySQL, not by the LiteSpeed Web Server process itself. No matter how fast LiteSpeed runs, it has no control over how fast a PHP script can run, or how much memory a PHP script will consume. Hence, in general, troubleshooting server load issue is not a matter of LiteSpeed Web Server's configuration, and is beyond our support scope. (Analysis of your DDoS attack is beyond our support scope as well, unless you would like to engage us through paid hourly support.)
Troubleshooting Tips¶
In the spirit of helpfulness, however, we are happy to provide the following troubleshooting tips for you to use when investigating high server load.
Try Caching¶
Try to use our LSCache solutions to lower the load.
Find CPU Hogs¶
It is always good practice to use the top
or atop
command to check which processes are using a lot of CPU resources, and then figure out whether they are behaving normally.
Check Disk I/O¶
If disk I/O is high, it is likely a disk I/O problem, which you should address. If you have more free memory available, you can move PHP sessions to /dev/shm
.
Example
top -c
iowait 41.6 wa
is too high. Check further:
iostat -x
Linux 3.10.0-962.3.2.lve1.5.24.10.el7.x86_64 (lbr27.lalunahost.com.br) 04/18/2019 _x86_64_ (8 CPU)
avg-cpu: %user %nice %system %iowait %steal %idle
6.65 0.15 2.16 20.24 0.00 70.80
Device: rrqm/s wrqm/s r/s w/s rkB/s wkB/s avgrq-sz avgqu-sz await r_await w_await svctm %util
sdb 0.16 7.43 16.61 1.80 74.74 78.56 16.65 2.46 133.39 8.38 1284.98 1.45 2.68
sda 28.25 134.39 93.98 54.24 1445.73 1227.69 36.07 2.11 14.18 46.67 79.17 4.86 72.08
You can see that /dev/sda
has a problem with 72.08%util
, and /dev/sdb
is ok.
Check CPU Idle¶
Check the CPU idle percentage. If it is close to 0%, then your server is CPU-bound and you should try to reduce the CPU usage.
Use CloudLinux LVE¶
When CloudLinux is used, you should try to use the LVE limit to prevent any one user from consuming all of the resources.
Upgrade PHP¶
Try PHP7 instead of PHP5 if your code is compatible with it.
Try Opcode Cache¶
Enable PHP opcode cache.
Check Real-Time Stats¶
Check the real-time stats for the number of PHP processes during peak time under Actions > Real-Time Stats > External Application. Look at In Use, Idle, and WaitQ. For shared hosting, if one user uses too many PHP processes while others only use a few, you may adjust PHP max concurrency setting to be a smaller number.
Are you Under Attack?¶
DDoS attack may push the server to high load. Check the number of concurrent users to see if it is normal:
netstat -na | grep 80 | grep ESTA
To check concurrent connections sorted by IP, run the following:
netstat -ntu | grep ESTABLISHED | awk '{print $5}' | cut -d: -f1 | sort | uniq -c | sort -nr
Sometimes you may count time_waits
as well by running the command without grep ESTABLISHED
, like so:
netstat -ntu | awk '{print $5}' | cut -d: -f1 | sort | uniq -c | sort -nr
It is useful in some special situations (i.e., an attacker could make a connection and then send requests to an expensive URL, wait a little while, and then close the connection. If the server does not abort the process, the backend will be used up soon and keep serving a request that has been abandoned. Hence, you wait to analyze IP even with time_waits
).
You can try some of the LSWS WAF features to mitigate DDoS attacks. If you can identify one or two IPs causing the problem, you should find a way to limit the resources that one or two bad IPs can consume. The web server has no control over how fast a PHP script can run, or how much memory a PHP script will consume. Reasonably, LVE limit by CloudLinux can help when someone abuses one account. If a server is under attack or being abused, you need to find a way to stop the attack or abuse, such as blocking bad IPs with a blacklist or using a modg_security rule set. No matter how fast a web server can run, it can only handle the amount of traffic the backend (PHP and MySQL) can handle, unless content is served from cache.
Check PHP Versions¶
Check your PHP configuration and make sure Apache and LiteSpeed have matching PHPs.
Disable open_basedir
¶
Disable the PHP open_basedir
feature.
Check the Code¶
Sometimes high load can be caused by poorly written PHP/SQL code and has to be fixed by your PHP Developers. Sometimes it is a MySQL-based problem, for example, you may have a missing index. To debug this, you may try to use mysqladmin
and processlist
. If you have spare RAM, trying to raise database cache pool/buffer may help. No web server configuration change can help such situations and they are beyond our support scope.
You can even use strace
to track your PHP processes: strace -tt -T -p <pid_of_a_php_process>
top - 15:28:56 up 1 day, 17:58, 3 users, load average: 187.42, 143.52, 109.09
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
25713 mysql 20 0 15.3g 784m 9.9m S 197.1 0.9 29:13.90 /usr/sbin/mysqld --basedir=/usr -- datadir=/var/lib/mysql --plugin-
30260 roan24pl 20 0 302m 103m 12m R 34.4 0.1 1:37.47 lsphp:/home/roan24pl/public_html/index.php
30448 roan24pl 20 0 302m 103m 12m R 26.0 0.1 1:09.17 lsphp:/home/roan24pl/public_html/index.php
30217 roan24pl 20 0 302m 102m 12m R 25.0 0.1 1:27.36 lsphp:/home/roan24pl/public_html/index.php
28755 roan24pl 20 0 291m 92m 12m R 24.0 0.1 6:52.65 lsphp:/home/roan24pl/public_html/index.php
Try Memcached¶
If you share sessions via MySQL database, you can try using memcached instead, to reduce the amount of MySQL queries.
Consider More Workers¶
More workers on your LiteSpeed license may mean better performance and reduced CPU load, but it depends on what your traffic looks like. If it is mainly PHP and MySQL processed contributing to server load, adding an LSWS worker won't help.
It is only when LiteSpeed worker processes' CPU usage is high, like over 50%, that you should consider adding more workers. Usually, this will happen when the server needs to process large amounts of HTTPS traffic with static content.
How many cores do you have? Generally your server is not considered overloaded until the load reaches the number of cores.
Conclusion¶
Addressing the above may lower your server load. However, each situation may vary greatly. We recommend ordering our LiteSpeed premium support service for a professional review, fine tuning, and installation/enabling of necessary elements to reduce your server load and maximize site performance.