From: Miller Alan (Alan.Miller@is-energy.de)
Date: Mon Oct 06 2003 - 07:09:06 EDT
I'm seeing less-than-optimal performance on all my GE interfaces of my SF15K.
Using ttcp I'm sending (from host-A) 1GB to a recieving host (host-C) as follows:
[root@host-A] rsh host-C ttcp -r -v
[root@host-A] ttcp -t -v -n131072 -D host-C
Host-A is a SF15K Domain, Host-C is an IBM 7026-6M1. Both on the same IP subnet.
Numerous tests (various times of the day/weekdays/weekend) show raw throughput
rates of 260 to 400 Mbps.
When I run the same tests from Host-B (another IBM, same subnet) to a Host-C
I'm seeing raw throughput rates between 460 and 660 Mbps.
I noticed a recent Sun-Alert (56641) stating that certain GE Cards may run at 33MHZ on the
PCI bis instead of 66 MHz and that in such cases, the max attainable bandwith is
Between 400 and 500 Mbps.
My cards are all running at 33MHz but I'm not even getting over 400Mps.
Prtdiag:
--------
Bus Max
IO Port Bus Freq Bus Dev,
Slot ID Type ID Side MHz Freq Func State Name Model
---------- ---- ---- ---- ---- ---- ---- ----- -------------------------------- ----------------------
/IO00/C3V0 PCI 28 A 33 66 1,0 ok network-pci108e,2bad SUNW,pci-gem
/IO01/C3V0 PCI 60 A 33 66 1,0 ok network-pci108e,2bad SUNW,pci-gem
/IO02/C3V0 PCI 92 A 33 66 1,0 ok network-pci108e,2bad SUNW,pci-gem
/IO03/C3V0 PCI 124 A 33 66 1,0 ok network-pci108e,2bad SUNW,pci-gem
/IO04/C3V0 PCI 156 A 33 66 1,0 ok network-pci108e,2bad SUNW,pci-gem
Now my questions:
Does anybody know how they came to that calculation?
Isn't the theoretical throughput as follows?
64 (bit bus) * 33 (MHZ) == 2.1 Gbits/second
echo "64 * 33000000"|bc
2112000000
How much throughput can I realisticly expect to get?
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:27:14 EDT