IBM pSeries manual Dropped switch packets

Page 24

MPCI: sends = 14

MPCI: sendsComplete = 14

MPCI: sendWaitsComplete = 17

MPCI: recvs = 17

MPCI: recvWaitsComplete = 13

MPCI: earlyArrivals = 5

MPCI: earlyArrivalsMatched = 5

MPCI: lateArrivals = 8

MPCI: shoves = 10

MPCI: pulls = 13

MPCI: threadedLockYields = 0

MPCI: unorderedMsgs = 0

LAPI: Tot_dup_pkt_cnt=0

LAPI: Tot_retrans_pkt_cnt=0

LAPI: Tot_gho_pkt_cnt=0

LAPI: Tot_pkt_sent_cnt=14

LAPI: Tot_pkt_recv_cnt=15

LAPI: Tot_data_sent=4194

LAPI: Tot_data_recv=3511

5.12 Dropped switch packets

Lower than expected performance can be caused by dropped packets on the HPS switch. Packets sent over a switch interface can be dropped in several ways, as described in the following sections.

5.12.1Packets dropped because of a software problem on an endpoint

Packets are sometimes dropped at one of the endpoints of the packet transfer. In this case, you should be able to run AIX 5L commands to see some evidence on the endpoint that dropped the packet. For example, run /usr/sni/sni.snap -l {adapter_number} to get the correct endpoint data. This is best taken both before and after re-creating the problem. The sni.snap creates a new archive in /var/adm/sni/snaps. For example, /usr/sni/sni.snap -

l1 produces a hostname.adapter_no.timestamp file such as /var/adm/sni/snaps/c704f2n01.1.041118122825.FEFE5.sni.snap.tar.Z.

For IP traffic, looking at netstat -Ddata is a good place to start:

netstat -D

 

 

 

 

Source

Ipkts

Opkts

Idrops

Odrops

[. . .]

 

 

 

 

sn_dmx0

156495

N/A

0

N/A

sn_dmx1

243602

N/A

0

N/A

sn_if0

156495

140693

0

0

sn_if1

243602

241028

0

13

ml_if0

0

98

0

10

The ifsn_dump command provides interface-layer statistics for the sni interfaces. This tool helps you diagnose packet drops seen in netstat -Dand also prints some drops that are not shown under netstat.

pshpstuningguidewp040105.doc

Page 24

Image 24
Contents IBM ~pSeries High Performance Switch Contents Mpprintenv Mpstatistics Introduction Tunables and settings for switch software MPI tunables for Parallel EnvironmentMpeagerlimit MppollingintervalMprexmitbufsize and Mprexmitbufcnt MemoryaffinityMptaskaffinity MpcssinterruptMPI-IO Chgsni command Tunables and settings for AIX 5L IP tunablesFile cache Svmon and vmstat commands Svmon Vsid Esid Type Description LPage Inuse Pin Pgsp VirtualVmstat Pin Pgsp VirtualLarge page sizing Pshpstuningguidewp040105.doc Large pages and IP support Memory affinity for a single LparAmount of memory available Debug settings in the AIX 5L kernel Daemon configurationRsct daemons LoadLeveler daemons Reducing the number of daemons runningReducing logging Settings for AIX 5L threads Placement of POE managers and LoadLeveler schedulerAIX 5L mail, spool, and sync daemons Debug settings and data collection tools Lsattr tuningDriverdebug setting Iptrclvl settingSmall Real Mode Address Region on HMC GUI Deconfigured L3 cacheService focal point Affinity LPARsErrpt command HMC error loggingMultiple versions of MPI libraries Mpprintenv Mpstatistics MemoryaffinityDropped switch packets Nddipacketsmsw 0x00000000 Nddipacketslsw Packets dropped in the ML0 interface Packets dropped because of a hardware problem on an endpoint Packets dropped in the switch hardware MpinfolevelLapidebugcommtimeout LapidebugperfAIX 5L trace for daemon activity Conclusions and summaryAdditional reading HPS documentationPOWER4 MPI documentationAIX 5L performance guides IBM RedbooksPshpstuningguidewp040105.doc