You right, it is just an isolated node.
We don't have to take this into account
Hal Rosenstock a écrit :
On Mon, Nov 23, 2009 at 5:09 AM, Céline Bourde <[email protected]> wrote:
My /var/log/opensm.log output:
Nov 20 14:33:58 824870 [9C8566F0] 0x02 -> osm_vendor_init: 1000 pending
umads specified
Nov 20 14:33:58 825157 [9C8566F0] 0x80 -> Entering DISCOVERING state
Using default GUID 0x2c9000100d00056d
Loading Cached Option:qos_max_vls = 8
Loading Cached Option:qos_high_limit = 1
Loading Cached Option:qos_vlarb_high = 0:1,1:0,2:0,3:0,4:0
Loading Cached Option:qos_vlarb_low = 0:1,1:2,2:4,3:8,4:0,5:32
Loading Cached Option:qos_sl2vl = 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
Entering MASTER state
Nov 20 14:33:58 840416 [9C8566F0] 0x02 -> osm_vendor_bind: Binding to port
0x2c9000100d00056d
Nov 20 14:33:58 877496 [9C8566F0] 0x02 -> osm_vendor_bind: Binding to port
0x2c9000100d00056d
Nov 20 14:33:58 877646 [9C8566F0] 0x02 -> osm_vendor_bind: Binding to port
0x2c9000100d00056d
Nov 20 14:33:58 877794 [9C8566F0] 0x02 -> osm_opensm_bind: Setting IS_SM on
port 0x2c9000100d00056d
Nov 20 14:33:58 952249 [98C4F910] 0x80 -> Entering MASTER state
Nov 20 14:33:58 952467 [98C4F910] 0x02 -> osm_qos_parse_policy_file: Loading
QoS policy file (/etc/opensm/qos-policy.conf)
Nov 20 14:33:59 047411 [9824E910] 0x01 -> sm_mad_ctrl_rcv_callback: ERR
3111: Error status = 0x1C
Nov 20 14:33:59 047894 [9824E910] 0x01 -> SMP dump:
base_ver................0x1
mgmt_class..............0x81
class_ver...............0x1
method..................0x81 (SubnGetResp)
D bit...................0x1
status..................0x1C
hop_ptr.................0x0
hop_count...............0x2
trans_id................0x18ef
attr_id.................0x17
(SLtoVLMappingTable)
resv....................0x0
attr_mod................0x0
m_key...................0x0000000000000000
dr_slid.................65535
dr_dlid.................65535
Initial path: 0,1,31
Return path: 0,22,1
Reserved: [0][0][0][0][0][0][0]
01 23 45 67 01 23 45 6F 00 00 00 00 00 00 00
00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00
What node (NodeInfo) is at direct route 0,1,31 relative to the SM node ?
Is it just this AM failing for the SLtoVLMapping attribute or others too ?
-- Hal
Céline Bourde a écrit :
Hi,
I'm trying to configure qos-policy.donc file for service-id use.
I've tried basic RDMA_PS_TCP service-id level with service id 0x0106
[]# cat /etc/opensm/qos-policy.conf
qos-levels
qos-level
name: DEFAULT
sl: 0
end-qos-level
qos-level
name: TCP
sl: 4
end-qos-level
qos-level
name: MPI
sl: 5
end-qos-level
end-qos-levels
qos-ulps
default : 0 # default SL
any, service-id 0x0000000001060000- 0x000000000106FFFF : 4
end-qos-ulps
I add this rule in my configuration and check by mapping
sl 4 on a vl with a weight of 0
# QoS default options
qos_max_vls 8
qos_high_limit 1
qos_vlarb_high 0:1,1:0,2:0,3:0,4:0
qos_vlarb_low 0:1,1:2,2:4,3:8,4:0,5:32
qos_sl2vl 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
I've launched qperf on the server side
and listen on client:
j=100; for i in $(seq 0 $j) ; do /usr/bin/qperf -ri mlx4_0:1 -li mlx4_0:1
10.12.1.4 -sl 4 -lp 20004 rc_bw; done;
rc_bw:
bw = 0 bytes/sec
rc_bw:
bw = 0 bytes/sec
j=100; for i in $(seq 0 $j) ; do /usr/bin/qperf -ri mlx4_0:1 -li mlx4_0:1
10.12.1.4 -sl 5 -lp 20005 rc_bw; done;
rc_bw:
bw = 3.37 GB/sec
rc_bw:
bw = 3.37 GB/sec
Qperf gives me results I expected due to qos-level configuration part, but
no
expected results using qperf tcp_bw, bandwith is not
filtered/blocked by sl weight.
# j=100; for i in $(seq 0 $j) ; do /usr/bin/qperf 10.12.1.4 tcp_bw; done;
tcp_bw:
bw = 923 MB/sec
tcp_bw:
bw = 935 MB/sec
j=100; for i in $(seq 0 $j) ; do /usr/bin/qperf -ri mlx4_0:1 -li mlx4_0:1
10.12.1.4 -sl 5 -lp 20005 rc_bw; done;
rc_bw:
bw = 2.23 GB/sec
rc_bw:
bw = 2.24 GB/sec
rc_bw:
bw = 2.21 GB/sec
Could you help me to understanding service-id mechanism or give me a
relevant
test to test TCP service id level ?
Thanks.
Céline Bourde.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
