iSCSI config HPE 5900AF 48XGT 4QSFP+ Switch (JG336A)?

[spgsitsupport]

Any special considerations/config for the use of iSCSI on this device?

Thanks

Seb

[spgsitsupport]

I done my best "guesswork" for this configuration, but the iSCSI access seems to be slower than my previous Dell N4024 stack

Output from one iSCSI dedicated port:

[HPE5900-SR1]dis interface Ten-GigabitEthernet 4/0/33
Ten-GigabitEthernet4/0/33
Current state: UP
Line protocol state: UP
IP packet frame type: Ethernet II, hardware address: 2c23-3abd-4648
Description: Ten-GigabitEthernet4/0/33 Interface
Bandwidth: 10000000 kbps
Loopback is not set
Media type is twisted pair
port hardware type is 10G_BASE_T
10Gbps-speed mode, full-duplex mode
Link speed type is autonegotiation, link duplex type is autonegotiation
Flow-control receive enabled
Maximum frame length: 10000
Allow jumbo frames to pass
Broadcast max-ratio: 100%
Multicast max-ratio: 100%
Unicast max-ratio: 100%
PVID: 77
MDI type: Automdix
Port link-type: Access
 Tagged VLANs:   None
 Untagged VLANs: 77
Port priority: 0
Last clearing of counters: Never
 Peak input rate: 67929207 bytes/sec, at 2016-10-24 19:05:23
 Peak output rate: 20791021 bytes/sec, at 2016-10-21 20:11:38
 Last 300 second input: 530 packets/sec 3342134 bytes/sec 0%
 Last 300 second output: 418 packets/sec 681748 bytes/sec 0%
 Input (total):  508503915 packets, 3395785841360 bytes
         508467805 unicasts, 0 broadcasts, 0 multicasts, 36110 pauses
 Input (normal):  508467805 packets, - bytes
         508467805 unicasts, 0 broadcasts, 0 multicasts, 36110 pauses
 Input:  0 input errors, 0 runts, 0 giants, 0 throttles
         0 CRC, 0 frame, - overruns, 0 aborts
         - ignored, - parity errors
 Output (total): 363930564 packets, 619066646134 bytes
         362466851 unicasts, 674021 broadcasts, 789692 multicasts, 0 pauses
 Output (normal): 363930564 packets, - bytes
         362466851 unicasts, 674021 broadcasts, 789692 multicasts, 0 pauses
 Output: 0 output errors, - underruns, - buffer failures
         0 aborts, 0 deferred, 0 collisions, 0 late collisions
         0 lost carrier, - no carrier

Anybody has any constructive input?

Thanks

Seb

[bamaskas]

Hello,

I am left guessing, given a view of just one idle iSCSI port, about a simple read or write performance comparison.  Are the initiator and target ports configured the same, using an access port VLAN 77 with receive  pause enabled ?  Perhaps ports should have flow-control (both transmit and receive) configured. Also check to ensure target jumbo frames are supported, assuming the initiators are also jumbo frame enabled.

[spgsitsupport]

Jumbo frames are configured (as always per iSCSI)

As to flow-control - probably this is the best I could find and should work with Intel X540 cards

And this is also worth reading

[bamaskas]

I am the author of the second paper you reference.  However you are not configuring DCB (lossless) iSCSI.  Just Ethernet pause, end to end.  I see that you reference Intel NICs as initiators.  Is 3PAR 8200 the iSCSI target or is it a 7200 iSCSI target? The 7200 does not support DCB iSCSI. 

[spgsitsupport]

No, my storage is plain Dell MD3600i dual controller 10Gb array

2 servers are fitted with two dual X540 Intel cards (round robin with subnet) via 2 untagged Vlans on 5900 IRF stack

At the moment my iSCSI ports on the 5900 are configured in rather simplistic way (same as I had on Dell N4032 stack):

interface Ten-GigabitEthernet3/0/33
port link-mode bridge
port access vlan 77
flow-control receive enable
stp edge-port

[spgsitsupport]

I have not run any actual speed test, but I can see that the access to storage from VMs machines that previously were perfectly fine just feels sluggish to me (and so does to my various users)

Seb

[spgsitsupport]

Had to revert to using Dell N4024 for iSCSI data network, as with 5900AF it was simply unusable

Probably some settings (or maybe a bug), but I have NO idea what.

On Dell I did not need to do anything very specific to get way better performance

Seb