Skip to main content
Unix & Linux

Return to Answer

replaced http://stackoverflow.com/ with https://stackoverflow.com/
Source Link
Community Bot
Community Bot
  • 1

It was the kernel. We were using 2.6.32, which is quite old, but it is the one supported by Red Hat EL and Scientific Linux.

Today I had lunch with a friend (Giuseppe OttavianoGiuseppe Ottaviano) who had a similar experience tuning high-performance indexing algorithms. After upgrading to the newest version everything (compiler, libraries, etc.) he changed the kernel to the 3.10 line and suddenly everything worked fine.

It worked for us, too. With the 3.10 kernel (courtesy of http://elrepo.org), all problems vanished.

Giuseppe suspects a pernicious interaction between NUMA and kernel paging, which leads to the kernel loading and saving like crazy the same data, slowing the machine almost up to a halt.

It was the kernel. We were using 2.6.32, which is quite old, but it is the one supported by Red Hat EL and Scientific Linux.

Today I had lunch with a friend (Giuseppe Ottaviano) who had a similar experience tuning high-performance indexing algorithms. After upgrading to the newest version everything (compiler, libraries, etc.) he changed the kernel to the 3.10 line and suddenly everything worked fine.

It worked for us, too. With the 3.10 kernel (courtesy of http://elrepo.org), all problems vanished.

Giuseppe suspects a pernicious interaction between NUMA and kernel paging, which leads to the kernel loading and saving like crazy the same data, slowing the machine almost up to a halt.

It was the kernel. We were using 2.6.32, which is quite old, but it is the one supported by Red Hat EL and Scientific Linux.

Today I had lunch with a friend (Giuseppe Ottaviano) who had a similar experience tuning high-performance indexing algorithms. After upgrading to the newest version everything (compiler, libraries, etc.) he changed the kernel to the 3.10 line and suddenly everything worked fine.

It worked for us, too. With the 3.10 kernel (courtesy of http://elrepo.org), all problems vanished.

Giuseppe suspects a pernicious interaction between NUMA and kernel paging, which leads to the kernel loading and saving like crazy the same data, slowing the machine almost up to a halt.

Source Link
seba
seba
  • 251
  • 1
  • 2
  • 6

It was the kernel. We were using 2.6.32, which is quite old, but it is the one supported by Red Hat EL and Scientific Linux.

Today I had lunch with a friend (Giuseppe Ottaviano) who had a similar experience tuning high-performance indexing algorithms. After upgrading to the newest version everything (compiler, libraries, etc.) he changed the kernel to the 3.10 line and suddenly everything worked fine.

It worked for us, too. With the 3.10 kernel (courtesy of http://elrepo.org), all problems vanished.

Giuseppe suspects a pernicious interaction between NUMA and kernel paging, which leads to the kernel loading and saving like crazy the same data, slowing the machine almost up to a halt.