Red Hat Bugzilla – Bug 70787
i686 kernels slows down to unusable state after scp >2GB file from other box
Last modified: 2007-04-18 12:45:14 EDT
From Bugzilla Helper:
User-Agent: Mozilla/5.0 Galeon/1.2.5 (X11; Linux i686; U;) Gecko/0
Description of problem:
When copying large (>2GB) file via scp everything goes fine until last byte is
copied. Then on box where is copying performed kernel slow down immediately to
unusable state. Executing ls or who take about minute. Most of time CPUs spend
in system mode. Each process in R state add 1 to load (naturally). There are no
process in D state. Shutting down this system via reboot takes hours.
No messages in syslog.
Entire file is copied, but scp doesn't ends.
I found, that some kernels have not this problem:
But these have it:
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. other box$ dd if=/dev/zero of=2aG count=4000001; scp 2aG unstable:
2. wait on unstable until copying ends
Actual Results: scp doesn't end
system immediately slow down and become unusable
Expected Results: normal scp ending
no additional loads and system calls
Hardware: HP NetServer 6000 U3 - 4 Pentium III (Cascades) 700 MHz 2 MB cache, 4
GB RAM, AMI MegaRAID (typ T7, 32 MB RAM), two RAID5 volumes, HP SureStore
Ultrium 230, CD-ROM, Intel Ethernet Pro 100 NIC.
I have not second such server for testing and on "normal" PC boxes this problem
I have this problem on second HP NetServer 6000 U3 too.
I have this problem also with 2.4.18-10.i686.
I have the exact same problem on a machine with the 2.4.18-18.7.xsmp kernel.
the current erratum kernel has a new rmap VM patch to fix this
Yes. It works on kernel-smp-2.4.20-20.7.i686. Many thanks.