| ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange |
I looked over the rosbag code. There is function called startReadingVersion200 in bag.cpp that reads in chunk headers and connection index records. The chunk headers are spaced at 1MB interval in the file. So for a 30GB file you have 30000 chunks. I'm guessing that reading each chunk is causing a seek, so with a 5ms seek time reading the file would take around 2min. Which matches what I have seeing. This would explain why it loads quickly when reloading. Because the header don't take up that much space and are probably being buffered by the OS.
The load time could be improved if the chunk headers could be processed as they are encountered in the bag file. I'm not familiar with the bag format so I don't know how easy or hard that would be.
Note that this is still an issue: https://github.com/ros/ros_comm/issues/117
I'm still having this issue. I have to replay a 270Gb bagfile and iotop shows painfully slow disk reads:
Actual DISK READ: 35.90 M/s | Actual DISK WRITE: 0.00 B/s
17726 be/4 root 17.11 M/s 0.00 B/s 0.00 % 3.69 % play --queue 100 --rate 1.0 --delay 2.0 --start 0.0 2016-02-12-11-15-41.bag
In ROS Groovy, rqt_bag seems pretty capable of dealing with large bagfiles. Even a 75GB file opened in in a little under a minute.
In my case the playback starts fast (less than a second). I think that rosbag don't read the whole bagfile because I'm using a bagfile of 7.5GB I think it may depend on the configuration of your computer. I have a core i7 with 4GB RAM and a NVIDIA GTS 450. Hugs.
Your local hard disk would be SSD then, if it takes less than a second to play 7.5 GB file. For 123 GB bag file, it took 1 hour 30 minutes to play from HDD, while took less than 10 seconds to play from SSD.
I dont think there is GPU utilization in playing bag file, its not even multicored.
Asked: 2011-08-08 07:38:41 -0500
Seen: 3,613 times
Last updated: Mar 12 '13
ROS Answers is licensed under Creative Commons Attribution 3.0 Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.
