FC vs. iSCSI performance with Mac OS X Server

Question

Hi All,
I've relied on direct-attached Fiber Channel storage for our network server volumes since the early days of the Xserve RAID units. We use our network volumes for graphic design needs (with large files) over a Gigabit network, with some Windows clients, too. We're currently running Mac OS X 10.5.8 Server on an Xserve quad core (2008 model) with a 16 TB Enhance Technology UltraStor RS16FS single controller RAID unit with 4Gbit FC. It's worked extremely well for us, but it's going on 3 years old now, and I'm looking to replace it with a dual controller RAID unit for added piece of mind, either from Enhance Technology, Promise, or another vendor. I'll also be increasing the capacity to 32 TB.
The big question is whether I should make the leap to iSCSI with a new RAID unit? I understand the advantages of iSCSI, but I'm concerned about its real-world performance and how taxing it would be on the server itself. Any decrease in performance from our existing 4Gbit FC would be unacceptable. I'm also concerned about the learning curve involved in setting up an iSCSI SAN, since my networking skills are fairly basic -- I really don't get into the advanced functions of our managed switches, and the one time I tried to set up link aggregation, things went horribly wrong! Direct-attached FC sounds a lot simpler.
I'd get a PCIe 4-port Gigabit Ethernet card from Small Tree and use our existing SMC TigerStack II managed Gigabit switch to aggregate the ports, unless anyone thinks a new dedicated switch would be better. I'd likely buy the ATTO iSCSI initiator to make everything happen. I suppose if I wanted to put the old FC RAID unit on the SAN, I could use the
Any opinions, suggestions, or links to benchmarks would be appreciated.
Thanks

Answer 1

Thanks for the reply, MrHoffman -- I appreciate your insights. Just to be clear, I'm currently using a 4Gbit FC RAID system directly connected to an Xserve with an Apple FC card, so there's no SAN.
Just to be clear, there's a SAN here. You just don't have a SAN switch, based on your description.
I'm sharing out the volumes on that RAID unit via AFP and SMB over the GbE network, so that's really where the bottleneck exists
That's typical, and why I pointed to the GbE as the bottleneck. (If you listen very carefully to the server, you can hear the little skidding sounds as each of the SAN packets decelerates onto the GbE.)
Ultimately, I would like to create a small SAN, if only to allow my backup server direct access to the main network volumes for D2D backups, and potentially to allow certain client computers direct access to the volumes. It would be far easier to do with iSCSI than with FC, since as you said, I could use my existing network infrastructure.
That's a PCI-X or PCIe-class Mac, the Xsan software, a switch, and another connection into your existing (yes, you have one) SAN. As for the array, the prices on those range from Not Too Exorbitant to Oh My Aching Wallet. Used gear (where you can find it) can be a decent investment when you're on a budget.
Xsan has a fixed price, and the SAN switches tend to show up on the used market; they're common in the enterprise space, and a 4 Gb SAN switch is not even remotely new gear.
If you need Big Cheap Storage, then Direct Attached Storage (DAS) approach will be your cheapest option. (You almost have a DAS configuration now.) A PCI-X or PCIe controller or a RAID controller connected out to a Big Dumb Disk Array, err, a JBOD, or into a Big Not-So-Dumb RAID Array. If you have a controller or an open slot in your Xserve box.
I'm leaning towards maybe just adding another 16-drive JBOD unit to the existing FC system to meet our short-term storage needs. That doesn't address my desire to get a dual controller unit. But really, how often do RAID controllers fail?
Um, how often? Usually only when you have no current backups and a deadline, in my experience.