Storage Network Unification
by Hu Yoshida on Aug 10, 2008
The move to network unification recently took a step forward with an FCoE Test Drive Event in June 2008. FCoE can send FC protocols over an Ethernet network and makes it possible to have a common Ethernet network support FC storage and Ethernet message traffic. This would greatly reduce connectivity costs, especially for servers. Today servers require redundant connections for storage, LAN traffic, and clustering connectivity. A redundant pair of FC host Bus Adapters alone could cost as much as the server itself.
When one thinks of Storage and Ethernet, there is a concern about guaranteed delivery and quality of service. While LAN traffic can tolerate dropped packets and variable service, FC block storage expects packets to be delivered in order and within a limited window of time. Since disk storage is a rotating media, if a sector address is missed, the disk takes another spin before we have another shot at the sector, and eventually the I/O times out. In order for FCoE to support block storage it must overcome these deficiencies. Extensions are being made to Ethernet standards, called Convergence Enhanced Ethernet or CEE. CEE will provide for congestion management, elimination of dropped packets, ensure interoperability, priority processing, and packet scheduling. These standards are expected to be finalized in 2009.
Since the biggest advantage of convergence is for the servers, they are expected to be the first to move to FCoE with a Converged Network Adapter, CNA, combing a FC HBA with an Ethernet NIC card. Qlogic has a CNA that is ready to ship and CISCO has an Ethernet switch available now. Brocade will also have an Ethernet switch shortly. On the storage side, adoption is expected to be slower due the investment that data centers have in FC and the fact that FC works. It is expected that Ethernet switches will initially provide a bridge to FC storage. Moving to FCoE for storage arrays would only require the replacement of the front end port modules.
Chris Mellor points out that, SAN virtualization products, like EMC Invista and IBM SVC which reside in the FC SAN will require significant effort to convert to FCoE. Since HDS’s storage virtualization is done in the USP V storage controller, and not in the SAN Fabric, it is expected to be much easier for HDS to convert to FCoE.
One more advantage for Storage Virtualization in the Storage Controller.
Comments (4 )
“While LAN traffic can tolerate dropped packets and variable service, FC block storage expects packets to be delivered in order and within a limited window of time.”
Not all LAN traffic can tolerate dropped packets. That which can’t uses TCP, which also guarantees in order delivery. It’s been working for 25 years.
“Since disk storage is a rotating media, if a sector address is missed, the disk takes another spin before we have another shot at the sector, and eventually the I/O times out.”
So that means your system won’t work with an iSCSI interface, since there might be a re-transmit and that would cause a time-out? I imagine it does work, and you simply re-try the I/O.
Really this line of thought has occurred before. In the 1980s IBM campaigned Token Ring as better than Ethernet using exactly the deterministic response argument you make here. People screamed that Ethernet was unreliable and non-deterministic and thus couldn’t be used in important applications.
In the 1990s the same old argument was trotted out again for ATM, this time claiming voice, video, and multi-media needed deterministic, synchronized transfer times. Fortunes were made betting on ATM, but in the end it found the dust bin as well.
Ethernet prevailed again.
Now the storage world doesn’t want to give up its baroque networking standard. And what line do they choose? Yup, Ethernet isn’t deterministic. They say, we need a special media because plain old Ethernet isn’t good enough, and even better, we need a new networking stack because mature TCP/IP isn’t good enough either. Just as before fortunes will be made from those that buy this line, and just as before, the equipment will find the dust bin.
And Ethernet will prevail again.
Because good enough, is just that.
Thanks for your comments. Anyone else have a comment?
Is good enough, just that?
I wonder why iSCSI is mentioned in every article I read with FCoE. I would agree FCoE is a better technology but as what cost. I am seeing FCoE as Data Center with the expected price tag. iSCSI is for the most part free, with TOE options to scale, but not designed for the data center. I do not believe the cost or deployment concerns even compare. Now as far as Ethernet which is included in every server you buy today, that is very comparable and also runs iSCSI. FCoE will not be included and will require large budget commitments. For Ethernet it really is good enough.
[...] [HDS] “Moving to FCoE for storage arrays would only require the replacement of the front end port modules….it is expected to be much easier for HDS to convert to FCoE.” from Hu Yoshida’s latest blog (http://blogs.hds.com/hu/2008/08/storage_network_unification.html). [...]