ISCSI finally got its day in the sun last week. Dell’s acquisition of EqualLogic confirming what we all know -- people are drowning in their data. What does an outmoded technology do when the industry giants (like Dell) start casting their vote?
Fibre Channel is to iSCSI what Beta is to BlueRay. The Fibre Channel guys are running scared. Customers see the tremendous savings offered by iSCSI, not only in acquisition, training, and operational costs, but savings from advanced virtualization technologies like thin provisioning, resource aggregation and high availability; It’s about doing more with less and solutions that allow storage to be managed independently of the hardware in the same manner as the new matriarch of storage--VMware.
Fibre Channel vendors have seen the writing on the wall and are now trying to save their ship by hopping on the Ethernet bandwagon. The technology is called FCoE (Fibre Channel over Ethernet.) Catchy name, huh? Which presumably will allow application severs to connect to Fibre Channel SANs over standard Ethernet, helping reduce the need for costly HBAs and switches. The hyped advantage is the elimination of TCP/IP overhead.
Fibre Channel sheep (blind followers) have been positioning iSCSI as unviable since the day iSCSI it was conceived because of TCP/IP overhead and its impact on performance. Bahhh. Today’s server processors can easily handle the small incremental overhead of iSCSI TCP/IP and it’s only getting better. Intel’s new I/OAT technology offloads the TCP/IP stack onto one of the cores of a multicore CPU, allowing the applications to run unimpeded on the remaining cores. Not only do software initiators work perfectly running on today’s dual and quad-core processors, it will only get better with octal-cores. What the Fibre Channel guys won’t tell you is the important features you lose by not having TCP/IP: congestion control and guaranteed delivery.
Is iSCSI the be-all and end-all? Where does LeftHand see the market going? This is material for many future blog posts, but I’ll give you a few hints. We see the use of 10 Gig Ethernet and beyond allowing customers to create different “channels” of communications. One could be for the interprocess communications between blades on a blade server, another channel for Remote Direct Memory Access (RDMA) or Infiniband for high speed, low latency communications, and another for iSCSI and other IP-oriented connections. Yes and conceivably FCoE too, if it comes to that.
The last remaining question in a Fibre Channel mind is “what about performance?” I’m asked this question from customers just about every day, and my answer is simple. Performance has more to do with the SAN technology and architecture than it has to do with the protocol. Take LeftHand’s scalable SAN architecture for example. LeftHand virtualizes, load balances and aggregates the connections on each iSCSI array. For example, if you have 2 iSCSI arrays, with 2 gigabit ports on each array, you have an aggregate bandwidth to your SAN of 4Gbps, equivalent to 4Gbps Fibre Channel. Once you expand beyond 2 disk shelves, the systems exceed the available bandwidth of a single Fibre Channel link, and scales linear from there.
In our 100TB SAN validation test (www.lefthandnetworks.com/library/third_party_reviews.php) we demonstrated the performance of a 15 node LeftHand SAN replicated across 2 sites. Not only did we exceed the maximum random IOPS of most high-end Fibre Channel SANs on the market today with over 50,000, but we also demonstrated the capability to stream 35,000 videos simultaneously, equivalent to over 1,800 MB/s sustained throughput. Let’s see EMC do that with one of their Clariions! Of course the misperception of iSCSI being slow still persists among those poor customers that were brainwashed “Symmetrix style.” Little do they know that LeftHand’s clustered iSCSI architecture can be scaled to not only outperform Symmetrix, but at the same time also offer higher availability at a much lower cost.
Why is Fibre Channel rushing to new solutions? Survival. Maybe the Fibre channel guys went running for cover when Bell Labs presented results that claimed serial 100-Gbit/s transmission across a 2,000km distance. Clearly with 10 Gigabit Ethernet gaining traction and 100 Gigabit Ethernet on the horizon, FCoE is Fibre Channel’s only hope of survival. As iSCSI continues to take market share, there may not be any share left to take by the time the Fibre Channel guys get done re-inventing TCP/IP and FCoE finally hits the market. By this time the TCP/IP issue will be buried in processor cores.