Comments

By Gary Stiehr, May 22, 2006 2:20 am

MIT is developing a petascale storage system composed of Hammer Z-Series storage elements that use Zetera’s Z-SAN technology. Each Hammer Z-Series can have up to 2 TB of SATA disk (offerings up to 9 TB later in 2006) and has a Gigabit Ethernet network interface. It is claimed by Hammer and Zetera that these storage elements can be added as needed to the network to scale up to 128 PB in one volume (Zetera also states this as “unlimited” scaling). The underlying storage fabric is IP and IP is used as the storage virtualization mechanism (via broadcasts and multicasts). Additionally, RAID functionality is provided in a distributed fashion by the Z-SAN technology included in each Hammer Z-Series unit.

As for performance, Zetera claims “an unlimited, linear increase in performance for each disk added to the fabric.” Hammer claims that each of its Z-Series units can provide up to 80 MB/s and up to “10+” GB/s across multiple enclosures. At 2 TB per enclosure and 80 MB/s per enclosure, you might expect to see 10 GB/s with 125 Z-Series. That would be 250 TB, so I am not sure why the figure (10+ GB/s) was chosen if the theoretical maximum volume size is 128 PB (which would require 64,000 Z-Series!). It seems like the potential aggregate transfer rate should be much higher.

Hammer states that their units are compatible with Red Hat Enterprise Linux 4 (Q2) and various Windows OSes. I guess then that a driver is provided by Hammer that communicates with the various Z-Series units. Zetera mentions that it uses UDP as well as “a packet-guarantee methods similar to TCP but with a bandwidth utilization of well over 90%.” According to Zetera, the potentially 128 PB volumes would appear and behave to the operating system as local disk drives.

With a volume of that size, I suppose one would choose to format it with XFS or Solaris ZFS. This could also be used as the backend storage to a global filesystem. On the other hand, Zetera mentions its Z-FS multi-initiator filesystem “delivers full file and volume sharing on block-level devices.” I am not quite sure if this provides a global filesystem capability or not.