From: Evan J. <ej...@uw...> - 2004-04-21 22:40:48
|
I was hoping that someone on this list might have some suggestions for what settings could be tweaked to prevent or to alleviate this problem: When I perform some operation on the UML that does a lot of disk writes (for example, copying a large file), the UML will become unresponsive until the write has completed. Running "vmstat 2" will pause, and when it returns it shows a huge number of block writes. Pings to the UML will have huge latency. I'm using an ext3 UML disk that has noatime set. The UML is running 2.4.24-2um. I'm guessing that the hang occurs because my UML is asking the host to "sync" a write to disk, and is hanging until that sync operation completes. Does that seem reasonable? Could using ext2 help? Tweaking the bdflush values? Any other ideas? Thank you for any assistance, Evan Jones Example: ej@ej:~$ time cp old-websites.tar.bz2 test.tar.bz2 [copy a 9MB file] real 0m0.301s user 0m0.000s sys 0m0.269s [ the operation itself completes VERY quickly, but immediately after the UML hangs ] In another console: ej@ej:~$ vmstat 2 procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu---- r b swpd free buff cache si so bi bo in cs us sy id wa 0 0 23228 3036 3184 25396 1 0 2 1 8 5 0 0 100 0 0 0 23228 3036 3184 25396 0 0 0 0 56 8 0 0 100 0 0 0 23228 3036 3184 25396 0 0 0 0 57 11 0 1 99 0 0 0 23228 3036 3184 25396 0 0 0 0 54 5 0 0 100 0 0 0 23228 1668 3272 26464 0 98 0 98 56 12 0 13 87 0 [ NO OUTPUT AS THE UML HANGS HERE FOR A MINUTE ] 3 0 23228 1612 3276 26464 0 0 0 3919 1835 7 0 99 1 0 0 0 23228 1580 3304 26500 0 0 19 41 54 19 20 6 74 0 0 0 23228 1576 3304 26500 0 0 0 0 54 2 0 0 100 0 0 0 23228 1572 3304 26500 0 0 0 0 56 4 0 0 100 0 0 0 23228 1572 3304 26500 0 0 0 0 55 2 0 0 100 0 0 0 23228 1572 3304 26500 0 0 0 0 54 5 0 0 100 0 From another machine: PING evanjones.ca: 56 data bytes 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=0. time=86. ms 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=1. time=86. ms 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=2. time=85. ms 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=3. time=85. ms 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=4. time=86. ms [ HUGE HANG ] 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=5. time=68608. ms 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=6. time=67609. ms 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=7. time=66609. ms 64 bytes from evanjones.ca (66.160.135.56): icmp_seq=8. time=65609. ms [ MANY MORE LINES OF DELAYED PING REPLIES ] -- Evan Jones: http://www.eng.uwaterloo.ca/~ejones/ "Computers are useless. They can only give answers" - Pablo Picasso |