There are three obvious ways you can deploy iSCSI. The most obvious way is natively, which basically is an all iSCSI SAN where all the devices have an Ethernet interface and connect to an Ethernet LAN and all use the iSCSI protocol without using any sort of translation or specialised equipment. This is the best way a small to mid-sized company could get into SANs.
Then there is bridging which is when you join Ethernet-based devices that are using the iSCSI protocol to an existing FC SAN. This deployment is typically carried out by organisations that have existing FC SANs and would like to migrate to an Ethernet based SAN. Specialised equipment has to be used here. We talk more about this later on.
The third way of using iSCSI, extension, is essentially linking SANs across large distances.
iSCSI considerations:
iSCSI is limited to current one gigabit Ethernet speeds while fibre channel can run at 2Gbps with 4Gbps just recently being made available. 10Gbps fibre is on the horizon, but so is 10Gbps Ethernet. For the time being, if you have already invested heavily in fibre think carefully before making an investment in iSCSI. At the moment the technology is slower, and over larger distances the delays will be even greater.
We know that FC works and it would be a safe bet to run mission critical systems over fibre. For now, we can't actually say this about iSCSI.
iSCSI vendors in our minds are having some doubts over large implementations, particularly because of the large delays that can occur. When 10Gbps Ethernet becomes easily available, we expect to see deployments of iSCSI becoming more common.
The Storage Network Industry Association (SNIA), who announced the iSCSI standard a few years back, says small to medium businesses will welcome this technology as a means to maximise their return on investment. Despite the big hype vendors have been somewhat slow in certifying their products. We are seeing this technology moving slowly but it is being used successfully by those who are keen to take advantage.
iSCSI Cards, TOEs and Standard NICs
You are probably wondering how your servers will connect using iSCSI. There are three ways: with a standard network interface card (NIC) using an iSCSI driver, with a TCP offload engine (TOE) NIC with an iSCSI driver, or by with an host bus adaptor (HBA) designed for iSCSI.
iSCSI places quite a large overhead on server CPUs. As iSCSI wraps additional information over standard SCSI commands, it creates a huge amount of TCP/IP work which a standard NIC offloads to the CPU. TOEs or HBAs can look after the overheads associated with iSCSI without offloading any work to the main CPU. The main difference between these two types of cards is iSCSI HBAs only handle iSCSI traffic whereas TOE cards handle all Ethernet traffic.
What will it cost?
Traditional SCSI DAS deployments have the lowest acquisition costs, however they bear much higher long-term costs, or TCO, than FC SAN and iSCSI SAN. Furthermore, FC SANs have a higher upfront cost that iSCSI SANs.
For this technology overview we take a first hand look at the Snap Appliance Server 4500 and how it implements iSCSI. We also look at some products from Adaptec, HP, and Cisco.
|
|
6%
1%
