Apparently, on May 15th MIT's Media Lab will announce a research project using huge amounts of storage to support "the most extensive scientific analysis of infant learning patterns every undertaken." The project uses very low-cost storage from a consortium of Bell Microproducts, Marvell, Seagate and Zetera to support petabytes of capacity.
It uses storage over IP Protocol (SOIP). The claim from Zetera is that SOIP will revolutionise storage just as voice over IP (VOIP) is revolutionising telephone communications. It is a variant of the 'all networking will become IP networking' concept.
Zeteras CEO, Chuck Cortright, said: "The promise of SoIP is scalable, high-performance network storage thats simple enough to be configured and managed by general IT staff." So that's storage management cost-savings. Then there are hardware and software cost savings: "The economic advantages of SoIP come from its elimination of expensive dedicated hardware in favour of cost-effective, mass-produced components.
Techworld has written about Zetera and its low-cost SAN idea before. Now Zetera appears to have a ground-breaking and high-visibility MIT project about to be publicise its Z-SAN technology.
IP addresses and partitions
The technical overview is simple enough. Servers with applications that need to access files and/or blocks access what they think is a local drive, such as G:. In fact they access a disk drive across and IP network. This disk drive has its own IP address.
In fact each partition on a disk could have its own IP address, becoming an IP Partition. IP partitions contain a range of logical block addresses (LBAs) on one or more disk drives. Using standard unicast and multicast network modes, IP partitions can be addressed individually or as groups. Groups will be presented to a client or clients as one or more logical volumes. Groups can contain from two to thousands of individual IP partitions.
By aggregating partitions i this way, Z-SAN technology can build large volumes with disks in different arrays, rooms or even cities. A single IP partition can be as large as 128 petabytes, meaning 1,000 terabytes, meaning one million gigabytes.
The IP addresses are accessible from any client device (servers) on the network. The disks are directly attached to the IP network linked to the servers, meaning Ethernet. A single drive can have multiple volumes acting, in effect, as several virtual disks.
An IP switch multicasts to the many disks (network nodes). Apparently there is then no need for expensive Fibre Channel SAN switches and directors and cables and HBAs. There is also no need for dedicated IP SAN components such as TOEs (TCP/IP offload engines). There is no need for dedicated NAS storage with its built-in NAS server or head. Neither is there any need for RAID hardware or specialised virtualisation hardware appliances.
Commodity disks are used with IP to create a peer-to-peer network in which storage is simply an extension of networking. Network bandwidth limitations are taken care of by 10 gigabit Ethernet. IP numbering limitations are taken care of by IP version 6 with its extended addressing scheme.
The storage network controlling intelligence
So, okay, storage paradise is here but there is one question. Certain functions still need carrying out. Granted servers can access there own disks because something, somewhere maps a drive letter to a disk's IP address but what 'thing':-
- Creates and manages volumes of aggregated disks and
- Protects data through striping and mirroring and
- Ensures network performance and
- Virtualises disks and
- Delivers files or blocks to servers?
There has to be some 'thing' carrying out these complex tasks.
Zetera says IP is inherently virtualised. Z-SAN technology uses standard UDP (User Datagram Protocol) which has a bandwidth utilisation of over 90 percent. (ISCSI is based on TCP/IP and sacrifices, Zetera says, 25 percent of its bandwidth to TCP/IP overhead.) Here is a Wikipedia note about UDP:-
"The User Datagram Protocol (UDP) is one of the core protocols of the Internet protocol suite. Using UDP, programs on networked computers can send short messages known as datagrams to one another. UDP does not provide the reliability and ordering guarantees that TCP does; datagrams may arrive out of order or go missing without notice. However, as a result, UDP is faster and more efficient for many lightweight or time-sensitive purposes. Also its stateless nature is useful for servers that answer small queries from huge numbers of clients."
"Common network applications that use UDP include the Domain Name System (DNS), streaming media applications, Voice over IP, Trivial File Transfer Protocol (TFTP), and online games."
Doesn't sound promising; datagrams may go missing without notice. But that is not a concern with the Zetera product. Ryan Malone, Zetera's senior marketing director, said: "UDP does not concern itself with guaranteed data delivery, leaving that responsibility to the application that sits above the UDP port. Zetera's Z-SAN technology absolutely guarantees delivery of each and every transfer in an atomic fashion. There is an ACK response for every transfer."
"If that transfer is lost (due to a lack of an ACK), then the protocol will re-issue the command. If commands continue to have no ACK, then the file system reports the error. Because UDP leaves responsibility of guaranteed data delivery to the application above the UDP port, Zetera has more flexibility and control over exceptions than TCP allows. Zetera's Z-SAN technology was built on a UDP/IP foundation with reliability protection that puts its data delivery reliability on par with Fibre Channel.
So no worries on that score at all.
Servers have a client driver to interpret file system commands and change them to block-level Z-SAN network messages. Each disk has a modular network controller or logical controller which receives these messages and converts them into disk-oriented I/O commands and/or payloads. There is a distributed controller architecture.
Between the accessing servers (or clients in Z-SAN speak) and the IP-addressable disks there is a standard IP switch. Each controller monitors the network traffic passing it, discards unwanted packets, detects and respons to packets meant for IP partitions on its disk drive.
The distributed controller architecture means that adding a disk/controller pair adds both capacity and performance to the system (but it also adds traffic to the network).
Z-FS file system
Z-SAN technology includes a multi-initiator file system enabling file and volume sharing on block-level disk drives. It is said to outperform conventional server and NAS file systems.
Data can be striped across multiple drives for faster read-write performance and better protection, better than RAID 5 for example. With Z-RAID stripes and mirrors can be combined on the same physical disk.
Zetera provides web-based management and provisioning tools and it provides the modular controller technology twinned with each disk drive. It appears to me that the job of the general IT admin staff is going to have to include setting up volumes, setting up IP partitions, setting up stripes and mirrors and so forth.
In large shops with thousands of drives this is going to be an onerous task. Thus it will be an SME-focused storage method if it becomes popular, one for file access, e-mail, etc. Don't expect it to be used for critical transaction processing systems. Storage commentators like Jon Toigo are very, very keen on SOIP.
The MIT project may help Zetera lot. Equally it may be dismissed as a marketing gimmick like the $100 computer. Watch this space.