how fast should iSCSI transfer be?

[chimoe125]

I have been trying to increase my iSCSI side of my network but am thinking I have some bottleneck somewhere.

I was doign a series of tests migrating vm from server local storage to iSCSI SAN and my avarage transfer rate is around 650MB per minute. Is this the avarage or below avarage transfer rate? This comes to around 10mb per second.

[golddiggie]

iSCSI performance depends on several factors. Network speed (between host and storage), storage ethernet controller speed, controller port count, controller configuration, physical drives inside the SAN, RAID level used inside the SAN on those drives, number (and speed) of the ports being used on the ESX/ESXi host servers, how the physical network is configured, to name just some. It could also be impacted if the bios/firmware of the SAN is out of date. Different storage arrays perform differently under different (or even the same) conditions. I know that one of the benefits of iSCSI storage from EqualLogic (such as the PS6000 series) is that you'll gain performance when you start stacking arrays together. Adding more arrays into a group increases the performance of the entire group, not just the amount of storage available to you.

It's often a good idea to keep an eye on the status of the pysical drives within the iSCSI SAN. If you have one that's faulty, and the RAID in a degraded state, that could impact the performance (not to mention other things).

Post up the make, model, disk count, disk type (and speeds) of your iSCSI SAN. Also what RAID level the drives are using.

VMware VCP4

Consider awarding points for "helpful" and/or "correct" answers.

[chimoe125]

Hello,

I understand there are lots of variables to what the speed would be but I'm trying to get an average to see if my 650MB per minute data transfer is normal.

I have standard NICs on servers that are 1GB (jumbo frames enabled)

Switch is Gigabit (Jumbo Frames enabled )

SAN has 2 Gigabit NICs and are SATA Drives (using RAID 1+0) (Jumbo frames enabled)

[geddam]

There is no bottom line writted as such of ISCSI transfer rate, as they depend on what type of storage you are using as backend.

However, These are test results I have for my setup...

Server1: Quad Core Xeon (Intel E5405) with 12GB of RAM and an Intel Pro 1000 NIC with HP Storage works

Gives 305,362kbs average speed with 1500 MTU (means no jumbo frames enabled), with MTU 9000, I have to test....

But there are many test cases available on web, however configuring it correctly and choosing the right hardware as always been the prime factor for ISCSI performance.

Thanks,,

Ramesh. Geddam,

VCP 3&4, MCTS(Hyper-V), SNIA SCP.

Please award points, if helpful

[amvmware]

You should see about 100 to 160MB/s if it was a lefthand or equallogic san.

You appear to have everything correct from a performance perspective - except SATA disks. They are OK for a test environment - i use them in mine - but for a production environment they are not going to deliver the level of IOPS required and is more than likely your bottle neck - your data transfer is only as fast as the slowest component. My undertanding is a 7.2K SATA disk will generate an average of 80 IOPS, a 5.4K SATA disk will generate 50 IOPS, compare to a 15K SCSI drive that can gnerate 150 - 180 IOPS and you see why you don't use SATA disks for production VM's.

Adding additional NIC's and using MPIO is probably not going to improve things significantly in this case.

[golddiggie]

What's the make and model of the iSCSI array you're using? Or what IS it? How many spindles does it have?

We had two EqualLogic iSCSI arrays stacked at my last job... They both contained 7200 rpm SATA drives (they were cheap buggers where I was last). Still, running <30 VM's off of the combined arrays saw no performance hit when compared to where they were before going virtual. Probably due to the 32 spindles (total) and how the EqualLogic arrays increase performance as you add more arrays into the group. The older of the two had dual ethernet controllers with three ports each. The newer had dual controllers with four ports each. ALL were Gb, on a dedicated Gb managed switch set for best possible performance. I never bothered to take IOPs numbers from the iSCSI arrays since performance was always in the acceptible range.

VMware VCP4

Consider awarding points for "helpful" and/or "correct" answers.

[RParker]

650 per minute seems a tad slow, considering 100-160 MB/s is the norm.

This comes to around 10mb per second.

This looks like a Fast Ethernet , which would be accurate given that 100MB Ethernet is ~ 10-12 MB/s