Page 1
    Page 2
  Page 3
   

 

 
By William Van Winkle
 
 
LAST MONTH, WE OPENED OUR storage discussion with a look at SATA and SAS drives and controllers as an SMB prospect. In particular, we delved into the differences between desktop SATA, enterprise SATA, and SAS hardware as well as some of the key elements in storage controllers. All of this was to provide a foundation leading to this month's conclusion: the storage box.

Predictably, the more money you pour into a storage solution, the more scalability, redundancy, and manageability you're likely to get. As a custom solution reseller, the challenge is to deliver as many of those advantages as possible for the fewest long-term dollars. Inevitably, the core of this pursuit revolves around the boxes that house all of the customer's storage and the connections that make that data manageable. In our spin around the industry, we've found some persuasive and exciting possibilities. Some of these you may already be involved with, but in all cases there's always room to deepen and expand your offerings.
 
 
Notes on NAS

At heart, network-attached storage (NAS) is a data storage repository that dangles directly off the LAN rather than functioning as direct-attached storage (DAS) for a specific system. NAS evolved from the mid-‘80s to early ‘90s principally as a means to liberate conventional UNIX servers from the cumbersome job of file serving. DAS often has an advantage on data throughput performance—consider 3 Gbps eSATA versus 1 Gbps Ethernet. However, NAS wins on its ability to function independently of any other machine. Just keep in mind that if part of the network fails, especially the closest switch or access point, the NAS appliance could be left stranded and effectively offline.

Network-attached storage is a file-level technology, meaning that files are transferred as a complete entity using protocols such as SMB or NFS, common on Windows and UNIX platforms respectively. Alternatively, block-level transfers move pieces of files and are operating system agnostic. Ignoring any sort of usage considerations, you could say that the difference between NAS and SAN (storage area networks) is one of file-level versus block-level transfers. This becomes significant when considering external storage for applications such as a SQL Server database. While you could use a NAS box for this, file-level transfer dictates that a file is unavailable while it is being saved, moved, or copied. When dealing with large numbers of users needing simultaneous access to a file or database, file-level transfer can kill realtime performance and even result in data loss. SAN dodges this problem by sending small pieces of data and not tying up an entire file.

One of the main functions of NAS is to lighten the demand load placed on servers, but you don't need an engineering degree to reason out that too many users hammering on a NAS box at once will bring the unit to its knees. What's less obvious is that there are several possible reasons for a NAS device's crippling. A NAS appliance is much like a PC in that it contains a motherboard, CPU, memory, hard disk(s), operating system, and at least one Ethernet port. A 750GB Seagate Barracuda 7200.10 has a sustained throughput of up to 78 MB/sec (modern SATA drives average about 50 to 70 MB/sec sustained), and you can improve this with RAID striping across multiple drives. A Gigabit port specs at 1,000 Mbps, or 125 MB/sec, so the drives would appear to be the bottleneck—but this usually isn't so.

We've seen reviews posting write speed results on Western Digital's dual-drive My Book World Edition II NAS device of under 3.5 MB/sec. The same test run on D-Link's DNS-323 dual-drive NAS enclosure hit over 7.5 MB/sec. Changing from the default spanning mode to RAID 0 striping with two drives yields only spotty improvements in single-digit percentages. Ultimately, the real bottleneck is the controller hardware behind the Ethernet port, and you can't always take price as a guide for performance. Because it's loaded with plenty of software value-adds, the Western Digital drive sells for $449. D-Link's empty enclosure streets for about $200, and a pair of 500GB drives adds another $270 to $300. Moreover, D-Link's product, being far easier to upgrade, is potentially more reseller-friendly.

Your World Made Easy
Western Digital's My My Book World Edition II (1TB) may not be the fastest NAS around, but bundled and optional apps make LAN-based and online file/folder sharing a breeze, even for newbies.

For an interesting comparison of low-end "toaster" NAS appliances, check here: www.smallnetbuilder.com/component/option,com_nas/Itemid,190. According to SmallNetBuilder's roundup of 30 different NAS solutions (which shows the DNS-323 scoring 8.8 MB/sec), only one product performed above the 12.5 MB/sec inherent in a 100 Mbps Fast Ethernet connection: Iomega's 320GB Storcenter Pro 200d (15.3 MB/sec). Again, a NAS box is essentially just a cheap, optimized PC, albeit without most of the external ports and internal slots. The speed of the storage controller and/or CPU is going to determine final performance. This is why you see a massive jump in speed when moving into higher-end NAS devices that use desktop-class processors and/or hardware-assisted storage controllers.

This raises an interesting question: Why not build and sell your own in-house NAS solution out of conventional PC components? By all means, if your customer isn't picky about aesthetics, you may end up rewarding a little learning with a lot of profit. We're hard pressed to think of a better use for all of those aging, high-gigahertz, single-core CPUs still sitting in inventory than fueling a four- or five-drive RAID 5 mini-tower for under $700. Most low-end NAS boxes use a customized Linux distribution, and there are some free options suitable for turning a PC into a NAS appliance. Chief among these is perhaps FreeNAS (www.freenas.org), which is still in beta. This is a software system based on the FreeBSD OS, a Web interface, and PHP scripts. According to the site, FreeNAS supports "CIFS (samba), FTP, NFS, AFP, RSYNC, iSCSI protocols, S.M.A.R.T., local user authentication, software RAID (0, 1, 5), [and] a full Web configuration interface."

Roll Your Own Storage Appliance
You don't have to pay extra for a sexy, pre-made NAS box. With ordinary PC components and a storage OS like FreeNAS, you can create solid storage solutions on the cheap.

A DIY NAS device will be a good fit for some but not others. Scalability may become an issue, for example. In the case of FreeNAS, hardware expansion is partially limited by OS compatibility. If you want to connect out to an external enclosure and add more drives, the RAID controller will have to support FreeBSD 6. The network adapter must also support FreeBSD 6. Most likely, you won't have hot swap drive support, and for businesses where footprints matter, a four-drive NAS like the Buffalo TeraStation Pro II or Intel's SS4000-E may more than compensate for its price premium and possible performance drop through optimized features and conveniences, including more stable and user-friendly management software.

Taking the DIY idea another step, you might ask, "So...if I take a dual-CPU Xeon or Opteron config and throw something like Windows Storage Server at it, isn't this also a NAS box? Or is it a storage server?" Now we're down to arguing philosophy and semantics. FreeNAS is comfortable booting from a 32MB CompactFlash or USB flash drive. You won't get the same hospitality from Windows Storage Server, TrueSAN's Cloudbreak, NetApp's ONTAP, or any similar OSes featuring far greater server capabilities than the humble FreeNAS. In the end, the difference between NAS and storage servers comes down to the ability to host applications.


The Big Three Connectors

By now, you know all about the asteroid impact of serial technology and the slow, global death of parallel. Parallel printer ports ceded way for Universal Serial Bus. More importantly for this discussion, parallel ATA is now evolving into SATA. And the lumbering giant, SCSI, only survives against Serial Attached SCSI (SAS) through the virtue of its expensive, decades-long install base. All of SCSI's best attributes have been rolled forward into newer interfaces; only the parallel trappings are being left for dead.

Two Interfaces, Same Drive
Seagate's 15,000 RPM Cheetah drive comes in both Fibre Channel (pictured here) and SAS varieties. While the drives are very similar, a company's infrastructure will decide which format is best.

We said before that the difference between NAS and SAN is file-level versus block-level transfer. The other way to look at the question is in terms of application. NAS is essentially an island of storage connected by a thin bridge to the server mainland. DAS follows the same concept, only without the LAN in the middle. As we'll see in a bit, there are ways to cascade, or daisy chain, NAS and DAS devices, but the larger an organization gets the more circumstances will arise when it makes sense to start making new bridges directly between those islands or even pushing them together into a larger landmass. This is the essence of a SAN. SAN is about building a fabric that weaves together multiple storage repositories into a more flexible yet cohesive whole.

In today's market, there are three primary connection technologies for tying together servers and storage repositories: Fibre Channel, iSCSI, and SAS.


[ FIBRE CHANNEL ]

The Fibre Channel (FC) concept was born in 1988 with the goal of simplifying the massive, high-speed enterprise data cables prevalent back then and extending their usable lengths. Originally, this length extension was achieved through the use of fiber optic cabling, and the technology was then called Fiber Channel. However, when copper conduit support was added to the specs, the British spelling (fibre) was adopted to help downplay any perception of exclusivity with fiber optics. Fibre Channel became an ANSI standard in 1994 and evolved into the de facto connection technology for SANs over the past decade. Part of the reason for FC's success was its agnosticism toward protocol support. SCSI is far and away the most commonly implemented protocol over Fibre, but IP, iSCSI, HiPPI, ATM, and others are all possibilities.

Truly
Robust RAID

While Adaptec recently consigned the SANbloc 2Gb Fibre Channel JBOD enclosure to its "legacy" lineup, the multi-level RAID box can pump out 350 MB/s and scale to 112 drives.

There are three types of FC topologies. Point-to-point is much like SATA or USB; it's just two Fibre devices connected to each other. A Fibre Channel arbitrated loop works along the lines of a spoked wheel, much like yesteryear's token ring. You can have up to 127 devices in an arbitrated loop, each connecting to the next, but if one goes down, you lose the entire chain. (Using FC hubs can help mitigate this problem.) Switched fabric topology allows for up to 16.7 million devices and is much like Ethernet in that devices are managed through FC switches from vendors such as ATTO, LSI, and QLogic.

The basic clock rate for Fibre is 1 Gbps, yielding an actual post-overhead throughput of about 100 MBps. The most recent FC spec, finalized in 2005, shows a line rate of 8.5 Gbps and actual throughput of 800 MBps, although 4 Gbps FC remains today's standard. (Work on a 10 Gbps spec is now under way.) Obviously, this compares favorably against the 300 MBps of modern SAS and SATA, but the comparative cost premium and learning curve entailed in a Fibre fabric can be substantial.


...more
 
         
    Back to top
Page   1 2 3
   
   
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form.