0

On one of my servers there are 173 [migration] processes listed in ps aux.

It has 8 processors so there are the regular 1 per processor right at the top of the list.

PID COMMAND
 2  [migration/0/0]
 4  [migration/0/1]
 6  [migration/0/2]
 8  [migration/0/3]
10  [migration/0/4]
12  [migration/0/5]
14  [migration/0/6]
16  [migration/0/7]

But then there are extras in this kind of format:

PID   COMMAND
 899  [migration/553/0]
 900  [migration/553/1]
2234  [migration/551/0]
2235  [migration/551/1]
8783  [migration/515/0]
8784  [migration/515/1]

The first number after migration seems to always be above 500 and less than 600, and they are always in pairs with the second number being /0 or /1.

There is also a scattering of other ones that look like this:

PID   COMMAND
411870 [migration/32006]
411871 [migration/32006]

There is no second number in the name, the first number seems to always be around 32,000 and the PIDs are massive.

Most of the processes were started in 2011 but 3 were started on January 4th and 14 were started on January 8th.

I found this because the server ran out of memory without any reason I could locate. It is a dedicated web server and it has a nearly identical (same cpu, mobo, etc...) database server next to it which has only 8 [migration] processes. I'm primarily wondering if this migration stuff is related or if this is an issue at all.

Nick
  • 287
  • 1
  • 10

2 Answers2

2

Turns out it is CloudLinux. Each isolated account gets its own migration processes. That confused me for some time. Hopefully someone else will find this useful as well.

Nick
  • 287
  • 1
  • 10
  • So this is not a dedicated webserver? This CloudLinux (what is that?) is based on CentOS 5.0? – Nils Jan 10 '12 at 10:07
  • You should (and are encouraged to) mark this answer as correct by clicking the tick beside this post. This lets future visitors to the site know this is the solution that worked for you. – Ben Pilbrow Jan 10 '12 at 20:23
  • Nils it is a dedicated webserver. CloudLinux is a kernel module and Apache module which isolates resources between different virtual hosts on a web server. Ben I would but it seems that there is a time limit on how quickly you can mark your own answer as the answer. :) – Nick Jan 10 '12 at 21:22
1

According to this these are kernel-threads for moving threads between CPUs. What is the system load of this system and how many threads are running on it?

Nils
  • 7,695
  • 3
  • 34
  • 73
  • The load on the system ranges from about 0.1 to 2.0. It averages 0.27. I'm not sure how to get a thread count, but the process count is on average around 350. Thats including the 170+ migration processes. – Nick Jan 09 '12 at 00:03
  • I just checked on different linux boxes that are currently not doing much - the number of migration-processes there is equal to the number of logicial CPUs. Each process is bound to its processor number: migration/2 is bound to CPU#2 and so on. I do not have a second "/" on those processes. I checked with SLES9 and CentOS 5. What is your operating system? – Nils Jan 09 '12 at 16:34
  • Operating system is CentOS running 2.6.18 in this case. I too thought that they were CPU bound and the first 8 are. Its the others that make no sense. – Nick Jan 09 '12 at 19:52