User Tools

Site Tools


nndocs:srp

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision
Previous revision
nndocs:srp [2025/02/24 18:25] naptasticnndocs:srp [2025/04/15 18:42] (current) naptastic
Line 5: Line 5:
   * targets are formatted like ib.fe80000000000000f4521403002c18b1   * targets are formatted like ib.fe80000000000000f4521403002c18b1
   * idk how you get those numbers except to examine an ipoib link, such as:   * idk how you get those numbers except to examine an ipoib link, such as:
- 
-  (reformatted slightly) 
  
   5: ib0: <...> mtu 2044 ...   5: ib0: <...> mtu 2044 ...
Line 13: Line 11:
  
 Now strip the first 4 bytes off (they change anyway) and remove the :'s Now strip the first 4 bytes off (they change anyway) and remove the :'s
 +
 +  fe800000000000005849560e59150301
 +
 +The top 64 bits of the ACL might need to be 0 or they might be the same as the GUID; I don't know the initiator ID actually gets generated and I'm not doing any more digging into it.
 +
 +  /srpt> ib.fe800000000000005849560e53b70b01/acls create ib.00000000000000005849560e59150301
 +  Created Node ACL for ib.00000000000000005849560e59150301
 +  Created mapped LUN 0.
 +
 +A Linux SRP target is always visible from all InfiniBand partitions. That limits its usefulness to me.
  
 ====Dependencies==== ====Dependencies====
Line 18: Line 26:
   apt install srptools   apt install srptools
  
-===Login fails===+Find targets to connect to:
  
-shark (initiator):+    # srp_daemon -o -v -c -p 1
  
-   [74794.509035] scsi host11: ib_srp: REJ received +  * -o means "run once" otherwise dmesg on all your hosts will get polluted with SRP login noise. 
-   [74794.509038] scsi host11: ib_srp: SRP LOGIN from fe80:0000:0000:0000:5849:560e:5915:0301 to fe80:0000:0000:0000:5849:560e:53b7:0b09 REJECTED, reason 0x00010001 +  * -v means "say what you're doing" 
-   [74794.509048] scsi host11: ib_srp: Connection 0/12 to fe80:0000:0000:0000:5849:560e:53b7:0b09 failed+  * -c means "emit target information in a format we can use later" 
 +  * -p 1 means "only scan on HCA port 1" so obviously change this if you are initiating from port 2...
  
-southpark (target):+Lazy benchmarking seems good:
  
-   [4483481.918835ib_srpt Received SRP_LOGIN_REQ with i_port_id 5849:560e:53b7:0b09:5849:560e:5915:0301, t_port_id 5849:560e:53b7:0b01:5849:560e:53b7:0b01 and it_iu_len 8260 on port 1 (guid=fe80:0000:0000:0000:5849:560e:53b7:0b09); pkey 0xb068 +  [root]@[shark][~]# dd if=/dev/sdb of=/dev/null bs=4M 
-   [4483481.945664] ib_srpt rejected SRP_LOGIN_REQ because target port ibp33s0f1_1 has not yet been enabled +  4096+0 records in 
-   [4483481.959142] ib_srpt Rejecting login with reason 0x10001+  4096+0 records out 
 +  17179869184 bytes (17 GB, 16 GiB) copied, 5.38771 s, 3.2 GB/s 
 +   
 +  [root]@[shark][~]# dd if=/dev/zero of=/dev/sdb bs=4M 
 +  dderror writing '/dev/sdb'No space left on device 
 +  4097+0 records in 
 +  4096+0 records out 
 +  17179869184 bytes (17 GB, 16 GiBcopied, 13.7431 s, 1.3 GB/s
  
 +...so I'm using some for swap. We'll see how long it takes me to induce a kernel panic. ;)
nndocs/srp.1740421512.txt.gz · Last modified: 2025/02/24 18:25 by naptastic