iSCSI on the Mac

On the Microsoft Windows OS, iSCSI has reached a very mature level. On Linux, the Linux-iSCSI project has a refined iSCSI initiator. On the Mac, there are only a few vendors, and most are on their first generation iSCSI protocol drivers for Mac OS X Panther 10.3 and Mac OS X Tiger 10.4 OS. This is all going to change in the next several months.

Apple Computer has a SAN solution, compatible with ADIC’s StorNext File System called xSAN, a new disk paradigm that supports Mac OS X application only. In order to use the xSAN file system, one needs to erase completely and reformat any attached drives, including RAID subsystems.

Currently xSAN is supported through fibre channel networks only, and requires a dedicated ethernet port for metadata. iSCSI allows a traditional gigabit ethernet (GbE) network to become dedicated to storage with a moderate performance decrease. One can install a dedicated single or multiple port PCI card for dedicated GbE storage networking and internet traffic.

ATTO Technologies offers the Xtend SAN and Studio Network Solutions offers the GlobalSAN product. In the near future, expect D-Link to have products ready for market that go beyond the 1 GbE connection, stretching the storage platform to 10 GbE, beyond the range of 4 Gb fibre channel.

Hive Mind Grid Computing Mac

There’s been a dream of mine to have multiple cheap machines networked in a way to form a hive mind, so commands on one distribute to all and work together. This is loosely called grid computing. Imagine 3 iMac G5’s, Power Mac G5’s, or PowerBook G4’s networked together to work as one. The future is now!

At NAB 2005, Apple announced Final Cut Studio, which allows distributed Compressor rendering/encoding, so through a dedicated gigabit ethernet or firewire infrastructure, you combine a few machines into a working super computer. Mac OS X Tiger has as it’s base a technology called Xgrid, which is a protocol for clustering machines together out of the box.

The secret to success is the gigabit ethernet connection running at 1 Gbps or 128 MB/s. so now in the new computing model, the CPU comes second to the network connection. In a few years 10 gigabit will be standard, passing 2 Gbps or 4 Gbps fibre channel.

So now that you can grid the machine together, how about storage? Apple has an Xsan solution. Xsan costs just below $1500 a workstation to get the computers to share storage ($999 Xsan seat license, $499 fibre channel PCI-X card). A much more elegant and economical solution, that works on panther 10.3.5 and above is iSCSI.

you remember SCSI? all macs had it before ATA. Storage engineers never gave up on SCSI, and are reintroducing the SCSI protocols, but over gigabit ethernet! On a more modern mac, you already have the port, or can easily get a $20 network PCI card (recommended approach to separate the storage network from the IP network). The beauty of iSCSI it can be entirely software based, so yes you get a performance hit, but imagine sharing all your hard drive on a small gigabit network using software SAN. This is revolutionary. Current vendors that support the Mac are ATTO technology and Studio Network Solutions. These companies are just the beginning to a wonderful marriage of storage, grid computing, and the Mac.

apple Xserve G5 cluster nodes
apple Xserve G5 cluster nodes

Networked Storage Basics

Almost since the beginning of desktop computing, the ability to share has been essential. Flash forward 30 years, and not only are computers networked, behold the pervasiveness of the internet to every workstation and laptop, but soon every computer will have shared storage. The simplest form of sharing would be networked attached storage (NAS), where you use built in networking protocols (TCP/IP, SAMBA, NFS, AppleTalk, etc.) to mount a remote volume over the internet or over a local area network (LAN) to a client system. You can now parcel files, using the file sharing aspects of the operating system.

A more robust system in the NAS space, uses an intelligent controller, or server (Mac OS X Server, NAS head, etc.) to deliver storage connectivity.

Companies like NetApp and BlueArc have mastered the front end of the storage array, and allow on the fly configuration, modeling, and allocation of storage resources as the needs of the enterprise change. Certain applications, like databases, require block level access to storage devices, as if they were directly attached to an individual computer. Application performance may suffer if the program is looking for block level access, but only sees NAS, however, in general, NAS is fast and is a completely viable storage platform.

Be it a NAS environment or a storage area network (SAN), most likely SAN is the storage subsystem architecture anyway. Currently there are three ways to attach a SAN to a workstation or workgroup:

Fibre Channel
Gigabit Ethernet, GbE (iSCSI)
IEEE 1394a or 1394b (FireWire)

Each methodology has is benefits and pitfalls, but all offer rugged access to shared storage resources. With a shared storage solution, the workstation can boot from the SAN, share files and applications (if the software licensing agreement allows it), and make the backup and restore aspect of data management easier.