At the SNIA event in 2007, we met some new storage realities, which are not so new, but certainly for the standard approaches we are used to with HP, IBM etc regarding the storage of large amounts of data in security and performance and scalability, it is a revolutionary approach.
The fundamental difference between reliable, fibre channel hardware storage with complex SAN architectures and large investments, compared to “storage clustering” can be summarized in a graph
Looking at the architecture of a storage cluster shown above, it is immediately clear why in storage clustering the throughput increases in proportion to the available space, while in the classic approach it remains constant. Classic storage systems generally have two redundant fibrechannel controllers and through the fiber network infrastructure (SAN) allow the use of disk space to all connected devices, as if they were physically mounted on the machines.
Thanks to the development of technologies such as the iSCSI protocol, more affordable storage realities have been developing. As is well known, a cluster is seen from the outside, by the rest of the network, as a single device through a virtual IP, but the real connectivity ports of every single small storage grow with the growth of the cluster hosts. That’s the reason for that graph.
At SNIA, we made contact with Isilon in the person of Gary Willoughby ( Isilon Sales Engineer) and Marcus Thompson, Regional Sales Director of LeftHand Networks.
The LeftHand SAN/iQ product was already known to us before and was taken by us as a reference.
Despite voices from storage specialists who underestimate these new approaches, we asked Isilon to evaluate an ISCSI storage cluster solution.
Our contact put us in touch with the very kind Xavier Guerin Country Manager, Southern Europe who provided us with an interesting estimate; The material subject to the estimate is listed below:
Software
3- OneFS Software License: Isilon IQ System Software
AutoBalance Software: Automatically balances data onto new Isilon IQ nodes
FlexProtect-AP Software: ECC+1, ECC +2 and mirroring
NFS Protocol: NFS file serving protocol for Unix/Linux clients
CIFS Protocol: CIFS file serving protocol for MS Windows clients
HTTP/WebDav Protocol: Reading and writing via HTTP/WebDav file serving protocol
FTP Protocol: FTP file serving protocol
NDMP Protocol: Backup and restore via NDMP protocol
Cluster Management Interface: Web GUI and command line cluster management
Systems
3- IQ 1920x Node: 2U node with 1.92TB capacity
36- 160GB SATA-2 disk drives
5.8- Total raw TB’s of capacity
12- Gigabytes of globally coherent cache
3- 2.0 GHz Dual-Core Intel Xeon Processor
6- Gigabit Ethernet front-side connectivity ports
6- InfiniBand intracluster HCA connectivity ports
European Union 250VAC Power Cords with CEE 7/7 Plug and IEC 60320 C13 Connector
Infiniband Switch
2- Flextronics InfiniBand 8-port Switch
Cables
6- 3 Meter Cable
The reason why such high-performance intracluster connectivity is necessary can be understood from the fact that storage clusters use a logic of redundancy and distribution of data between machines in order to obtain performance and reliability, as if we were building a RAID of disks, the concept is really very similar.