Messages in this thread | | | Date | Sat, 16 Oct 1999 20:11:01 +0200 | From | Kristian Koehntopp <> | Subject | Speeding up swap |
| |
A friend of mine has a machine which is being used for certain numerical problems. The system has 512 MB RAM and another 1.5 Gig of RAM. The memory objects in use are usually contiguous chunks of memory sized between 2 and 20 MB, which are normally accessed in a sequential manner.
Benchmarking the harddisk area which is being used for swap, he can get sustained write rates of 26 MB/sec under Windows NT 4.0SP5 as well as under Linux 2.2.12. Swapping a gigabyte of data takes 6 min under NT and 11 min under Linux, resulting in a data rate of 6 MB/sec under NT and 3 MB/sec under Linux. During paging operations heavy seeking is audible.
The questions are:
1. Why is this so slow compared to the benchmarked data rate?
2. What tools are available under Linux to determine the reason for the paging activity (i.e. are these only accesses for the desired memory objects or are we observing some strange interference between multiple processes?) or to show how memory objects map to swap area data blocks? Or, formulated more abstract, what tools are available to debug such problems?
3. Are there any system calls available which can be utilized to advise the system that a certain area of memory will always be accessed as a contigous chunk of memory that should be grouped contigously in a swap area if possible and to advise the system that all memory accesses to this area will be sequential/sequential backward/random so that the system can optimize its memory access strategy. Where would one search for documentation on such system calls?
Kristian
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |