This patch series add to the "C" based tools the final "major" functionality
of the script based tools. Furthermore, it marks the scripts deprecated and
creates a compat rpm (_not_ built by default) which contains those scripts for
those who, heaven forbid, scripted around those scripts. The motivation
behind this move is to make the overall functionality of the diags faster and
simpler.
The specific issues addressed are:
1) redundant functionality
ibcheckerrors vs ibqueryerrors
ibstat vs ibstatus
ibcheck[net|node] vs new iblinkinfo/ibqueryerrors "check" option
ibprint[ca|switch|rt] vs ib[hosts|switches|routers]
2) Additional functionality
iblinkinfo and ibqueryerrors can be configured on a link by
link basis for what links should be rather than having
ibcheck* scripts look for != 1X links and no check for speed
at all. Furthermore, support for partially populated fabrics
is enhanced, for example checking for links which should not
be.
3) Better Scalability.
iblinkinfo and ibqueryerrors can scan the fabric up to 30X
faster than the scripts. One example[*] I ran, compared
"ibqueryerrors -f" vs ibchecknet on my test nodes. My test
system includes just 2 switches and 2 nodes with 4 HCA's
(2/node). The time difference is staggering.
ibchecknet: 1.793 seconds
VS
ibqueryerrors: 0.063 seconds
This includes the additional functionality of ibqueryerrors!!!
4) remove outdated/deprecated scripts
perl versions of iblinkinfo/ibqueryerrors
5) Less dependence on GUID's
specifically ibfabricconf.xml files use node names rather than
node or port guids to resolve their checks. As large fabrics
are maintained CAs and switches are often swapped for
maintenance and names are a better abstraction for System
Admins rather than having to update GUID's in a config.[**]
One downside to this change is that there will not be any "check"
functionality which does not have to be configured. However, I don't believe
these "checks" were really useful to begin with. Each fabric is different and
the user has different expectations for how their fabric should be cabled and
what speed/width those cables should be.
For further information on the new ibfabricconf.xml support see the iblinkinfo
man page.
Ira
[**] Unfortunately, some hardware still exists which can not have node
descriptions programed into the firmware. Therefore the configuration of
NodeGUID to Node Name still must be configured.
[*] Here is an example of the different runs on my test system. 2 nodes, 2
switches, and 4 HCA's (2/node)
First the ibqueryerrors run:
bash-4.1# time ibqueryerrors -f
Errors for "happy HCA-3"
GUID 0x1175000079da38 port 1: [SymbolErrorCounter == 332]
[PortRcvErrors == 6]
Errors for 0x66a00e3003b39 "QLogic 12200 GUID=0x00066a00e3003b39"
GUID 0x66a00e3003b39 port ALL: [SymbolErrorCounter == 154]
[LinkDownedCounter == 2] [VL15Dropped == 23]
GUID 0x66a00e3003b39 port 35: [SymbolErrorCounter == 154]
[LinkDownedCounter == 1] [VL15Dropped == 23]
GUID 0x66a00e3003b39 port 36: [LinkDownedCounter == 1]
Errors for 0x2c9020040ff58 "Ira's Switch"
GUID 0x2c9020040ff58 port ALL: [LinkDownedCounter == 18]
[PortRcvSwitchRelayErrors == 6940] [PortXmitDiscards == 449] [PortXmitWait ==
560]
GUID 0x2c9020040ff58 port 19: [PortXmitDiscards == 8] [PortXmitWait
== 560]
GUID 0x2c9020040ff58 port 20: [LinkDownedCounter == 8]
[PortXmitDiscards == 1]
GUID 0x2c9020040ff58 port 21: [PortXmitDiscards == 230]
GUID 0x2c9020040ff58 port 22: [LinkDownedCounter == 4]
[PortRcvSwitchRelayErrors == 46] [PortXmitDiscards == 146]
GUID 0x2c9020040ff58 port 27: [PortRcvSwitchRelayErrors == 6894]
GUID 0x2c9020040ff58 port 35: [LinkDownedCounter == 4]
[PortXmitDiscards == 63]
GUID 0x2c9020040ff58 port 36: [LinkDownedCounter == 2]
[PortXmitDiscards == 1]
## Summary: 6 nodes checked, 3 bad nodes found
## 76 ports checked, 10 ports have errors beyond threshold
## Thresholds:
## Suppressed:
Reading fabric conf file...
Evaluating connectively...
ERR: port down: "QLogic 12200 GUID=0x00066a00e3003b39" p: 35[ ] <==>
p: 35 "Ira's Switch" (Should be: 4X 10.0 Gbps Active)
ERR: speed != 10.0 Gbps: "QLogic 12200 GUID=0x00066a00e3003b39" p: 36[
] <==(4X 5.0 Gbps Active/ LinkUp)==> p: 36[ ] "Ira's Switch" ( Could be 10.0
Gbps)
ERR: Unconfigured active link: "ending HCA-3" p: 2[ ] <==(4X 10.0
Gbps Active/ LinkUp)==> p: 20[ ] "Ira's Switch" ( )
ERR: port disabled: "Ira's Switch" p: 35[ ] <==> p: 35 "QLogic 12200
GUID=0x00066a00e3003b39" (Should be: 4X 10.0 Gbps Active)
Stats Summary: (76 total physical ports) 66 down ports(s) 1 disabled
ports(s)
5 link(s) at 4X
1 link(s) at 5.0 Gbps (DDR)
4 link(s) at 10.0 Gbps (QDR)
real 0m0.063s
user 0m0.003s
sys 0m0.015s
Now the ibchecknet run:
bash-4.1# time ./ibchecknet
#warn: counter SymbolErrorCounter = 154 (threshold 10) lid 10
port 255
Error check on lid 10 (QLogic 12200 GUID=0x00066a00e3003b39) port all:
FAILED
#warn: counter LinkDownedCounter = 18 (threshold 10) lid 4 port 255
#warn: counter PortRcvSwitchRelayErrors = 6940 (threshold 100) lid 4
port 255
#warn: counter PortXmitDiscards = 449 (threshold 100) lid 4 port 255
Error check on lid 4 (Infiniscale-IV Mellanox Technologies) port all:
FAILED
#warn: counter PortXmitDiscards = 230 (threshold 100) lid 4 port 21
Error check on lid 4 (Infiniscale-IV Mellanox Technologies) port 21:
FAILED
#warn: counter PortXmitDiscards = 146 (threshold 100) lid 4 port 22
Error check on lid 4 (Infiniscale-IV Mellanox Technologies) port 22:
FAILED
# Checking Ca: nodeguid 0x001175000079da38
#warn: counter SymbolErrorCounter = 332 (threshold 10) lid 5
port 1
Error check on lid 5 (happy HCA-3) port 1: FAILED
# Checking Ca: nodeguid 0x0002c90300108f2e
# Checking Ca: nodeguid 0x001175000077d90e
# Checking Ca: nodeguid 0x0002c903004bebda
*** WARNING ***: this command is deprecated; Please use "ibqueryerrors
-f"
## Summary: 6 nodes checked, 0 bad nodes found
## 10 ports checked, 0 bad ports found
## 3 ports have errors beyond threshold
real 0m1.793s
user 0m0.220s
sys 0m0.576s
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
