I am trying to use a Ceph kernel client to mount Ceph DFS(/) and set up a samba server to share the mount point(/mnt/) to Windows 7. when l write a large file to the shared directory ., I found that: Time eth0 HH:MM:SS KB/s in KB/s out CPU %user %nice %system %iowait %steal %idle 18:38:09 4540.85 199.65 06:38:09 PM all 4.48 0.00 9.45 17.41 0.00 68.66 18:38:10 108359.6 971.33 06:38:10 PM all 21.05 0.00 39.23 0.00 0.00 39.71 18:38:11 94209.18 837.11 06:38:11 PM all 27.00 0.00 54.00 0.00 0.00 19.00 18:38:12 108445.3 972.69 06:38:12 PM all 26.13 0.00 49.25 0.00 0.00 24.62 18:38:13 111292.8 976.45 06:38:13 PM all 29.65 0.00 47.74 2.51 0.00 20.10 18:38:14 110391.5 954.38 06:38:14 PM all 23.86 0.00 55.33 0.00 0.00 20.81 18:38:15 110909.2 972.39 06:38:15 PM all 28.50 0.00 55.00 0.00 0.00 16.50 18:38:16 105417.3 7160.54 06:38:16 PM all 28.14 0.00 53.27 0.00 0.00 18.59 18:38:17 93842.25 28704.94 06:38:17 PM all 17.50 0.00 52.50 9.50 0.00 20.50 18:38:18 54239.49 31787.57 06:38:18 PM all 22.50 0.00 41.50 0.00 0.00 36.00 18:38:19 34258.94 100884.5 06:38:19 PM all 18.09 0.00 66.83 2.51 0.00 12.56 18:38:20 31679.40 113713.2 06:38:20 PM all 10.95 0.00 89.05 0.00 0.00 0.00 18:38:21 49619.47 114486.8 06:38:21 PM all 11.62 0.00 78.28 0.00 0.00 10.10 18:38:22 40821.56 115698.2 06:38:22 PM all 12.50 0.00 72.50 0.00 0.00 15.00 18:38:23 47052.48 114899.3 06:38:23 PM all 11.94 0.00 75.12 0.00 0.00 12.94 18:38:24 37556.76 115517.8 06:38:24 PM all 14.43 0.00 66.67 0.00 0.00 18.91 18:38:25 59026.53 100313.4 06:38:25 PM all 11.00 0.00 74.00 0.00 0.00 15.00 18:38:26 65620.30 93774.74 06:38:26 PM all 14.14 0.00 74.24 1.01 0.00 10.61 18:38:27 102363.6 17002.45 06:38:27 PM all 22.50 0.00 60.00 0.00 0.00 17.50 18:38:28 110130.1 961.28 06:38:28 PM all 29.85 0.00 52.24 0.00 0.00 17.91 18:38:29 110177.4 974.45 06:38:29 PM all 29.65 0.00 51.26 0.50 0.00 18.59 18:38:30 111202.0 970.46 06:38:30 PM all 29.15 0.00 50.75 0.00 0.00 20.10 what happens is that the write performance starts to slow down horribly when the client write data to ceph, I'm wondering if this is an issue with pdflush or CPU performance limits. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html