RE: [openib-general] OFED-1.0-rc6 is available
The MTU change undos the changes for bug 81, so I have reopened bug 81 (http://openib.org/bugzilla/show_bug.cgi?id=81). With rc6, PCI-X osu_bw and osu_bibw performance is bad, and PCI-E osu_bibw performance is bad. I've enclosed some performance data, look at rc4 vs rc5 vs rc6 for Cougar/Cheetah/LionMini. Are there other benchmarks driving the changes in rc6 (and rc4)? Scott Weitzenkamp SQA and Release Manager Server Virtualization Business Unit Cisco Systems OSU MPI: · Added mpi_alltoall fine tuning parameters · Added default configuration/documentation file $MPIHOME/etc/mvapich.conf · Added shell configuration files $MPIHOME/etc/mvapich.csh , $MPIHOME/etc/mvapich.csh · Default MTU was changed back to 2K for InfiniHost III Ex and InfiniHost III Lx HCAs. For InfiniHost card recommended value is:VIADEV_DEFAULT_MTU=MTU1024 mpi_perf.xls Description: mpi_perf.xls ___ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Re: [openib-general] OFED-1.0-rc6 is available
Thanks... one further fix for Cisco gateways: sometimes the IsDM bit is set on switch ports as well, so ibsrpdm should not be limited to just CA ports. Here's the patch, also on the trunk as r7836. --- srptools/ChangeLog (revision 7803) +++ srptools/ChangeLog (working copy) @@ -1,3 +1,10 @@ +2006-06-08 Roland Dreier <[EMAIL PROTECTED]> + + * src/srp-dm.c (get_port_list): In some setups (eg Cisco SFS 3001 + with an FC gateway), there will be switches with the IsDM bit set + on port 0. So the initial get of NodeRecords must retrieve all + records, not just CA ports. + 2006-06-07 Roland Dreier <[EMAIL PROTECTED]> * src/srp-dm.c (do_port): Use correct endianness when comparing GUID against Topspin OUI. --- srptools/src/srp-dm.c (revision 7803) +++ srptools/src/srp-dm.c (working copy) @@ -523,11 +523,9 @@ static int get_port_list(int fd, uint32_ out_sa_mad->mgmt_class= SRP_MGMT_CLASS_SA; out_sa_mad->method= SRP_SA_METHOD_GET_TABLE; out_sa_mad->class_version = 2; - out_sa_mad->comp_mask = htonll(1ul << 4); /* node type */ + out_sa_mad->comp_mask = 0; /* Get all end ports */ out_sa_mad->rmpp_version = 1; out_sa_mad->rmpp_type = 1; - node = (void *) out_sa_mad->data; - node->type= 1; /* CA */ len = send_and_get(fd, &out_mad, in_mad, node_table_response_size); if (len < 0) ___ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
RE: [openib-general] OFED-1.0-rc6 is available
Roland did the fix on Trunk and I took it to OFED 1.0 branch. Tziporet -Original Message- From: Ramachandra K [mailto:[EMAIL PROTECTED] Sent: Wednesday, June 07, 2006 8:28 PM To: Tziporet Koren Cc: [EMAIL PROTECTED]; openib-general; Ramachandra K Subject: Re: [openib-general] OFED-1.0-rc6 is available Tziporet Koren wrote: > Hi All, > > We have prepared OFED 1.0 RC6. > From the openib source tar ball in OFED RC6, it looks like the SRP kernel changes (ulp/srp/ib_srp.c) in the trunk for supporting Rev 10 targets have been included in RC6, but the corresponding changes to the userspace srptool--ibsrpdm (userspace/srptools/src/srp-dm.c) for displaying the IO class of the target have not been made part of RC6. The changes to ibsrpdm were committed to the SVN repository trunk in revision number 7758. Will the latest version of ibsrpdm make it to the next OFED release ? Regards, Ram ___ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
RE: [openib-general] OFED-1.0-rc6 is available
We sure will. -Original Message- From: Roland Dreier [mailto:[EMAIL PROTECTED] Sent: Wednesday, June 07, 2006 11:03 PM To: Tziporet Koren; [EMAIL PROTECTED]; openib-general Subject: Re: [openib-general] OFED-1.0-rc6 is available We also just found a bug in how ibsrpdm discovers Cisco/Topspin FC gateways. The patch is below, and is also checked in to the trunk as svn rev 7803. Please include this in OFED 1.0 final. Thanks, Roland --- srptools/ChangeLog (revision 7796) +++ srptools/ChangeLog (working copy) @@ -1,3 +1,9 @@ +2006-06-07 Roland Dreier <[EMAIL PROTECTED]> + * src/srp-dm.c (do_port): Use correct endianness when comparing + GUID against Topspin OUI. + + * src/srp-dm.c (set_class_port_info): Trivial whitespace fixes. + 2006-05-29 Ishai Rabinovitz <[EMAIL PROTECTED]> * src/srp-dm.c (main): The agent ID array is declared with 0 --- srptools/src/srp-dm.c (revision 7796) +++ srptools/src/srp-dm.c (working copy) @@ -52,8 +52,6 @@ #include "ib_user_mad.h" #include "srp-dm.h" -static const uint8_t topspin_oui[3] = { 0x00, 0x05, 0xad }; - static char *umad_dev = "/dev/infiniband/umad0"; static char *port_sysfs_path; static int timeout_ms = 25000; @@ -249,7 +247,7 @@ static int set_class_port_info(int fd, u init_srp_dm_mad(&out_mad, agent[1], dlid, SRP_DM_ATTR_CLASS_PORT_INFO, 0); - out_dm_mad = (void *) out_mad.data; + out_dm_mad = (void *) out_mad.data; out_dm_mad->method = SRP_DM_METHOD_SET; cpi= (void *) out_dm_mad->data; @@ -266,9 +264,8 @@ static int set_class_port_info(int fd, u return -1; } - for (i = 0; i < 8; ++i) { + for (i = 0; i < 8; ++i) ((uint16_t *) cpi->trap_gid)[i] = htons(strtol(val + i * 5, NULL, 16)); - } if (send_and_get(fd, &out_mad, &in_mad, 0) < 0) return -1; @@ -371,7 +368,10 @@ static int do_port(int fd, uint32_t agen struct srp_dm_svc_entries svc_entries; int i, j, k; - if (!memcmp(&guid, topspin_oui, 3) && + static const uint64_t topspin_oui = 0x0005ad00ull; + static const uint64_t oui_mask= 0xff00ull; + + if ((guid & oui_mask) == topspin_oui && set_class_port_info(fd, agent, dlid)) fprintf(stderr, "Warning: set of ClassPortInfo failed\n"); ___ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Re: [openib-general] OFED-1.0-rc6 is available
We also just found a bug in how ibsrpdm discovers Cisco/Topspin FC gateways. The patch is below, and is also checked in to the trunk as svn rev 7803. Please include this in OFED 1.0 final. Thanks, Roland --- srptools/ChangeLog (revision 7796) +++ srptools/ChangeLog (working copy) @@ -1,3 +1,9 @@ +2006-06-07 Roland Dreier <[EMAIL PROTECTED]> + * src/srp-dm.c (do_port): Use correct endianness when comparing + GUID against Topspin OUI. + + * src/srp-dm.c (set_class_port_info): Trivial whitespace fixes. + 2006-05-29 Ishai Rabinovitz <[EMAIL PROTECTED]> * src/srp-dm.c (main): The agent ID array is declared with 0 --- srptools/src/srp-dm.c (revision 7796) +++ srptools/src/srp-dm.c (working copy) @@ -52,8 +52,6 @@ #include "ib_user_mad.h" #include "srp-dm.h" -static const uint8_t topspin_oui[3] = { 0x00, 0x05, 0xad }; - static char *umad_dev = "/dev/infiniband/umad0"; static char *port_sysfs_path; static int timeout_ms = 25000; @@ -249,7 +247,7 @@ static int set_class_port_info(int fd, u init_srp_dm_mad(&out_mad, agent[1], dlid, SRP_DM_ATTR_CLASS_PORT_INFO, 0); - out_dm_mad = (void *) out_mad.data; + out_dm_mad = (void *) out_mad.data; out_dm_mad->method = SRP_DM_METHOD_SET; cpi= (void *) out_dm_mad->data; @@ -266,9 +264,8 @@ static int set_class_port_info(int fd, u return -1; } - for (i = 0; i < 8; ++i) { + for (i = 0; i < 8; ++i) ((uint16_t *) cpi->trap_gid)[i] = htons(strtol(val + i * 5, NULL, 16)); - } if (send_and_get(fd, &out_mad, &in_mad, 0) < 0) return -1; @@ -371,7 +368,10 @@ static int do_port(int fd, uint32_t agen struct srp_dm_svc_entries svc_entries; int i, j, k; - if (!memcmp(&guid, topspin_oui, 3) && + static const uint64_t topspin_oui = 0x0005ad00ull; + static const uint64_t oui_mask= 0xff00ull; + + if ((guid & oui_mask) == topspin_oui && set_class_port_info(fd, agent, dlid)) fprintf(stderr, "Warning: set of ClassPortInfo failed\n"); ___ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
Re: [openib-general] OFED-1.0-rc6 is available
Tziporet Koren wrote: Hi All, We have prepared OFED 1.0 RC6. From the openib source tar ball in OFED RC6, it looks like the SRP kernel changes (ulp/srp/ib_srp.c) in the trunk for supporting Rev 10 targets have been included in RC6, but the corresponding changes to the userspace srptool--ibsrpdm (userspace/srptools/src/srp-dm.c) for displaying the IO class of the target have not been made part of RC6. The changes to ibsrpdm were committed to the SVN repository trunk in revision number 7758. Will the latest version of ibsrpdm make it to the next OFED release ? Regards, Ram ___ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general