Mirko_Huth
Enthusiast
Enthusiast

Poor iSCSI performance after vSphere upgrade

We are experience a iSCSI performance degradation since vSphere upgrade. I have not done exact measurements right now but its slower.

Has anyone similar issues or tips what to change in ESXi 4 to get the ESXi 3.5 iSCSI performance back?

Thanks!

Tags (1)
0 Kudos
6 Replies
paithal
VMware Employee
VMware Employee

Need more details on your test:

- What is the kind of test (workload) your are running to compare the results ?.

- What is the system model?.

- How many CPUs ?

- How many targets & LUNs you are using ?

- What is the target model & FW version ?

0 Kudos
Mirko_Huth
Enthusiast
Enthusiast

As said i have not measured the traffic but noticed that backup tasks (VCB / BE Agent Backup of files within the VM's) take longer as before the upgrade.

Here are some details of our cluster:

2 x Dell PE 2950 (identical systems with: Dual Quadcore 3 GHz, 24 GB RAM, 2 single port QLogic 4050c iSCSI HBA's, 1x Intel Quadport NIC)

2 x Dell PE 2900 (identical systems with: Dual Quadcore 3 GHz, 24 GB RAM, 2 single port QLogic 4050c iSCSI HBA's, 2 x Intel Quadport NIC's)

1 x Dell MD 3000i (2 x Dual Port iSCSI Controller)

1 x Dell MD 3000i (2 x Dual Port iSCSI Controller)

1x Dell MD 1000 (attached to one of the MD 3000i)

4 x Dell managed PowerConnect switches

Each QLogic HBA displays:

4 targets

9 Devices

22 Paths

The server share 5 data stores:

Datastore1: 36 VM's/Templates 1,09 TB

Datastore2: 3 VM's/Templates 196 GB

Datastore3: 1 VM 136 GB (RDM DB store for SQL Server running on Datastore1)

Datastore4: 1 VM 136 GB (RDM Log store for SQL Server running on Datastore1)

Two other LUN's on the SAN are used by our physikal Backup Server (2 TB / 1,2 TB)

We have two server rooms in different buildings. The vSphere cluster is split over both server rooms (1 x PE2900 / 1 x PE2950 / 1 x MD3000i / 2x Switch) in each room).

The switches connect the 4 ports of each md3000 to the HBA's providing full redundancy. One port of each switch is connected to a cisco stack that has multiple fiber connections to a stack in the other building.

The SAN traffic between the buildings is separated from the LAN by use of a VLAN. All ports in the SAN network use the same subnet.

The servers use the ESXi 4 installable version installed on local RAID 1 disks.

I can provide more details e.g. firmware when i'm in the office tomorrow.

Thanks!

Mirko

0 Kudos
kunhuang
Enthusiast
Enthusiast

can you check the fw version?

cat /proc/scsi/qla4xxx/?

it should be 3.0.1.49

for vsphere4.

you can download from qlogic website.

- Kun

0 Kudos
SGL3
Contributor
Contributor

Do the Qlogic HBA have ToE functionality? If yes can you try to disable it then check performance and see if it improves?

0 Kudos
kunhuang
Enthusiast
Enthusiast

I don't think ToE is relavent to the performance issue here. iSCSI hba offload tcpi/ip and iscsi stack altogether to hw adapter. The ToE is part of tcp which is conceptually a subset of tcp function, which is in turn offload in iscsi adapter.

0 Kudos
cbrou
Contributor
Contributor

I am curious as to what the fix ended up being for this. I am also experiencing slower iSCSI performance after a recent upgrade from ESX 3.5.

0 Kudos