[[!meta copyright="Copyright © 2010 Free Software Foundation, Inc."]] [[!meta license="""[[!toggle id="license" text="GFDL 1.2+"]][[!toggleable id="license" text="Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. A copy of the license is included in the section entitled [[GNU Free Documentation License|/fdl]]."]]"""]] [[!tag open_issue_gnumach]] IRC, unknown channel, unknown date: `cat < /dev/zero | cat > /dev/null` will eventually make the system crash, likewise when using a FIFO. <antrik> hm... VM activity seems much higher when running fifo than pfinet... may be the cause <antrik> "zero filled" and "page faults" are serveral times higher with pipe than with pfinet <antrik> (cow faults however are about the same...) <antrik> pflocal is about the same as fifo <antrik> no, because it usually takes like 20 minutes until it crashes, sometimes much longer <antrik> not sure, but the longest so far was in the range of hours IIRC <antrik> I think I never tested what happens on "cat /dev/zero >/dev/null"... another thing yet to try <antrik> Linux BTW seems to employ some major VM trickery in this case -- dd shows a transfer rate of 10 GB/s... <antrik> no, no anomalies in vmstat <antrik> the only observation I made is that number of page faults and some other number rise pretty quickly with pflocal and fifo, but not with pfinet <antrik> I guess that's somehow related to the fact that pfinet doesn't crash -- though I guess the difference is simply that pfinet is way slower... <antrik> (haven't checked that, though) <antrik> BTW, I'm not sure you got it right: the test case is "cat /dev/zero|cat >/dev/null", *not* "cat /dev/zero >/dev/null" <antrik> OK, "cat /dev/zero|tail -c 1" also crashes, so it's definitely not related to /dev/null <antrik> "dd if=/dev/zero|tail -c 1" crashes as well <antrik> but "tail -c 1 /dev/zero" doesn't seem to <antrik> cool... running multiple instances of the pipe test also considerably speeds up the crash