2008-04-01 12:24 PM
A LUN is a LUN is a LUN. Structurally, there are no differences between the two. The difference lays in the protocol which is used to access them. You can umap an FCP LUN and map it to an iSCSI initiator or vice versa.
In fact, with the ONTAP DSM 3.2 you can have a mix of iSCSI and FCP paths to the same LUN on a windows box.
2008-04-02 06:49 AM
The performance of an iSCSI LUN vs an FC LUN is dependent upon many things, the however, the structure of the LUN itself is not one of them. For example:
1) Are you using a dedicated network segment?
2) Is I/O to the LUN getting routed?
3) Do you use switches that share buffers among Port Groups?
4) Do you use Jumbo frames (end-to-end)
5) Possible driver issues on the host side. Is it possible the driver may not be responding to pause frame requests and continues to fwd packets which are dropped by the switch because of buffer overflow/overrun which requests a re-send of lost packets. Have you looked the switch port logs?
5) Have you try going Direct connect to the array Port to eliminate possible networking issues?
2008-04-02 12:56 PM
Just to give some figures:
we use a FAS3020c as our central log filer: all our databases use this system for there logs. Depending on the database type, we use iSCSI (ms SQL) or FCP (Sybase).
With our sybase, our response time (write) is less than 0.6 ms (between 0.5 and 0.6 ms). For iSCSI, we see a response time of 0.6 ms.
Conclusion: we see almost the same response time for both environments.
When you need throuput, than FCP has an advantage if you can't use 10 GbE for iSCSI.In the near future the price of an 10 GbE will drop and then you can use this more easy and will have a higher bandwith than FCP.
2008-04-02 01:07 PM
Are you using Hardware iSCSI HBA or TOE Card to get the same perfomance as of 2/4 GBS FC card??
How does the host see iscsi lun , is this same c0t0d0 format or in different format ??
2008-04-02 01:29 PM
iSCSI performance is one of the most misunderstood aspects of the protocol. Looking at it purely from a bandwidth perspective, Fibre Channel at 2/4Gbit certainly appears much faster than iSCSI at 1Gbit. However, there are two important terms that need to be defined: Bandwidth and Throughput
Bandwidth: The amount of data transferred over a specific time period. This is measured in KB/s, MB/s, GB/s
Throughput: The amount of work accomplished by the system over a specific time period. This is measured in IOPS (I/Os per second), TPS (transactions per second)
There is a significant difference between the two in that Throughput has varying I/O sizes which have a direct effect on Bandwidth. Consider an application that requires 5000 IOPS at a 4k block size. That translates to a bandwidth of 20MB/s. Now consider the same application but at a 64k size. That's a bandwidth of 320MB/s.
Naturally, as the I/O size increases the interconnect with the smaller bandwidth will become a bottleneck sooner than the interconnect with the larger one (iSCSI vs FC).
At small block random I/O (4k-8k) both protocols perform equally well with similar IOPs and latencies but as the I/O size increases one (iSCSI) gets affected more than the other (FC)
TOEs and iSCSI HBAs do not guarantee higher performance and that's not the reason to deploy them. In fact, for a lot of workloads the native SW intiators outperform the iSCSI HBAs. The reason these cards came to fruition was to offload TCP and iSCSI processing overhead from the CPU. In an already underutilized server they provide no benefit unless you want to boot and even then you still don't need them given that there are NICs out there that support IP SAN boot using native iscsi stacks.
2008-04-02 01:33 PM
No we don't use iSCSI HBA's, we have test this in the past (iSCSI HBA) but the microsoft iSCSI initiator is a great peace of software (for me, almost the best what microsoft has produced so far ).
So the performance is the same and in some cases better than an hardware card. The only disadvantage of the software initiator is that you consume some cpu cycles. But in the latest hardware, the CPU is not your bottleneck (even in a vmware environment) but your amount of memory is. So it's no problem to consume some cycles for iSCSI.
I have no (l)unix experiance with iSCSI. We use iSCSI only with windows. But I assume that there is also no difference between iSCSI and FCP.
2008-04-03 06:23 AM
No, there is no difference between a LUN served up via iSCS and a LUN served up via FCP. The difference is only in how the server connects to the LUN, the LUN is the same. As a matter of fact, I can create a database (SQL Server or Oracle or any other) in a LUN which is connected to the server via iSCSI. If I want to change this to FCP, I shut down the database, unmap the LUN from the iSCSI connection and remap it via an FCP connection and restart the database. The database or application will have no idea the connection protocol has just changed.
A LUN is a LUN, you connect to the LUN via iSCSI of FCP but the LUN doesn't change depending on which connection you choose.
Database Field Technical Lead