Jim Clarke that's Jim Clarke the founder of Silicon Graphics and not some other Jim Clarke you may have heard of was an ideas man and the company he founded was an ideas company. However Jim Clarke's Big Idea wasnt a particular item of technology, it was an integration of many concepts into a single product, a rendering pipeline.
And the customers loved it. Engineers designing a new product from scratch, animators creating eye-catching advertisements, earth scientists trying to visualise complex geological structures, they all bought SGI kit in their thousands. Not because it was the fastest, though it was certainly quick, but because when it came to 3D graphics it was simply the best of everything the most powerful software, the fastest rendering hardware, the biggest displays together in a single package.
Now the company is expanding its horizons beyond the 3D graphics market. Its InfiniteStorage range of storage products typify the same approach, integrating best-in-class hardware and software into a smoothly-functioning whole. Built around off-the-shelf components with a solid reputation for reliability switches from Brocade, disk drives from Engenio (the Disk-Manufacturer-Formerly-Known-As-LSI) and servers and filesystem software from SGI itself there's nothing remarkable or unconventional about the products, they just work very, very well.
Technically the InfiniteStorage range includes both simpler NAS products and full-blown SANs. The SAN 2000 sits roughly in the middle, in terms of cost, capacity and throughput. Equipped with a Brocade Silkworm switch, the SAN 2000 has 2 logical FC connections, allowing a throughput up to 400MBytes/sec. SGI's multi-O/S SAN filesystem, CXFS, comes as standard, along with XVM Volume Manager and Resource Manager software. Connection is supported to Windows, Mac OS X, Linux and Unix hosts, plus NFS, CIFS and Samba NAS units.
Disks are installed in trays of 14, either 146GB Fibre Channel or 250GB Serial ATA units, giving approximately 2TB per tray in the FC incarnation. With a tray of disks priced at approximately £12,000, raw disk space costs a modest 0.6p/MB.
It goes without saying the hardware has all the usual high-availability features RAID 5 as standard with RAID level changeable on the fly, redundant power supplies, all disk units hot-swappable, and so on. If space runs low extra disks can be installed while the unit is running and either added to an existing LUN or concatenated to the file system as a new volume. And in the event of more serious problems SGI offers 24x7 support with 2-hour response.
The modular approach makes products easily upgradeable. The basic NAS 2000 model becomes a SAN 2000 with the addition of a switch and appropriate software. Upgrade the switch and add two more FibreChannel connections and its the top-of-the-range SAN 3000. Thus storage capability can evolve as requirements change.
The real strength of SGI's offering is its CXFS filesystem. An extension of XFS, SGI's proprietary high performance 64-bit filesystem the 'C' stands for "clustered" CXFS is specifically designed to optimise SAN performance. In particular it provides flexible file sharing.
To understand why this is so important, think about a medium-to-large post-production shop. As soon as, say, newsreel footage is loaded into the system many different users need to access it, often in conflicting ways. A producer might want to scan the material for suitability, while a QC group is colour grading it, while an editor is incorporating it into a report, while But post-production schedules are not negotiable; yesterday's news is no news at all and time is, literally, money. No-one can afford to wait on others to finish what they're doing.
This is the classic workflow bind and too often the answer is multiple copies.
CXFS elegantly solves this problem with file-locking mechanisms, allowing multiple simultaneous accesses to a single file while still providing SAN-level performance.
While file data resides on the fast FC disks, metadata information describing the files, such as size, location, access restrictions etc. is stored on a separate metadata server with its own dedicated TCP/IP network, built-in failover, transaction logging (journaling) and rapid recovery if a failure does occur.
If a workflow involved much creating and deleting of small files it's conceivable disk I/O could bottleneck on metadata access, with the actual data service stalling as a result. CXFS goes to great pains to maintain metadata throughput, using fast algorithms and lookups, allocating disk space in large chunks to minimise requests and buffering wherever possible, but nonetheless the SAN 2000 is not going to be your first choice as a Web server.
On the other hand, in situations where huge quantities of data flow through a system in relatively few files, constantly accessed simultaneously from multiple sources, the SAN 2000 could be a very good storage server indeed.
Where huge quantities of data flow through a system in relatively few files, constantly accessed simultaneously from multiple sources, the SAN 2000 could be a very good storage server indeed.