| Blogs | Classifieds | Downloads | FlashChat | Gallery | Googlemap | Invite Friends | Links | Projects | Reviews | Wiki |
| |||||||||
Welcome to the pSeries Tech Forums,
our free peer-based support site for administrators, engineers and architects working with IBM pSeries servers and software. You are currently viewing our site as a guest which gives you limited access to view most discussions, articles, tutorials and access our other free features. By joining our community you will be able to collaborate with administrators, engineers and architects charged with designing, delivering or maintaining IBM pSeries server environments. Founded by a recognized IBM pSeries consultant and IBM Redbook author, pSeries Tech Forums was developed with the single mission of bringing IBM pSeries professionals together into a single self-help community. Registration is fast, simple and absolutely free to all IT professionals with responsibility for or interest in IBM pSeries servers. We invite you to join our community today! If you have any problems with the registration process or your account login, please contact contact support. |
| Our Sponsors | |
| | |
| Want to advertise? | |
![]() |
| | LinkBack | Thread Tools |
|
#1
| ||||
| ||||
Two lpars use virtual ethernet adapters to transfer data in a p595 system, I think they can send faster then the real ethernet adaper. I found some samples in the redbook "IBM System p Advanced POWER Virtualization Best Practices", the main point is tune the MTU, tcp_sendspace and tcp_receivespace parameters, but I can not match the effect using those combinations. such as MTU=65394, tcp_sendspace=262144, tcp_receivespace=262144 I use ftp to test with a big file over 2G, the transfer speed only 100MB/S, but it should be more than 600MB/S in the sample of page 84 to page 85. How can I do that? Or I have to do some configurations on virtual ethernet adapters? |
|
#2
| ||||
| ||||
I've ssen other tests here where people configure it and the data still goes over a physical adapter (or two). The majority of performance tuning on Virtual Ethernets you should reserve for PURELY virtual ethernets - ones with no external ports. You can then whack up the MTU size etc, but that only helps if you are transferring large chuncks of data. If you are transferring small amounts of data it will have no effect.
__________________ Ross Mather, IBM AIX IT Specialist. That said anything I say here is my own opinion and not anything that you can ever hold against IBM. Ohhh and don't forget that I make mistakes too.... |
|
#3
| ||||
| ||||
change cpu from 2 to 0.5 at the vio client, test throughput is 250MB/S, it's slower than 2 cpus normally. But when it write to disk, the speed is still about 80MB/S. If I use 0.4 cpu, than throughput is 180MB/S, write to disk is 89MB/S. If I use 0.2 cpu, than throughput is 90MB/S, write to disk is 45MB/S. I think 0.4-0.5 cpu can make the best performance with limited resources. The major bottolneck is the disk's read-write speed, and how can I tuning the disk speed ? Default LTG is 256. |
|
#4
| ||||
| ||||
Are you using the logical volume manager on the Virtual I/O server? Then you need the extra CPU for the LVM calculations there. I always advise against using the LVM on VIO where performance may be an issue. Walso what version of the VIO are you using?
__________________ Ross Mather, IBM AIX IT Specialist. That said anything I say here is my own opinion and not anything that you can ever hold against IBM. Ohhh and don't forget that I make mistakes too.... |
![]() |
| Bookmarks |
| Currently Active Users Viewing This Thread: 1 (0 members and 1 guests) | |
| Thread Tools | |
| |