STORAGE VS. SERVER

Sure, Intel has a server/workstation product group called “Storage Systems,” and Microsoft has a Windows OS called “Storage Server.” But names can be tricky things. Sometimes they work in your favor, and sometimes they don’t.

Take the SSR212MC2, a.k.a “McKay Creek,” from Intel’s storage systems lineup. This is a 2U, dual-processor system with twelve 3.5” drive bays up front, and it’s commonly called a storage server. When a customer shows up wanting a storage server for his 5TB data archive and tells you he needs lots of horsepower, the knee-jerk reaction may be to reach for a “storage server.” Blindly reach for an SSR212MC2, and you might be picking the perfect product. Or you could be overselling your way out of a bid...or underselling your way into future headaches.

Similarly, a customer might say he wants a streaming video server. The SSR212MC2 may well be the ideal machine for the job, but you pass it over. That’s a storage server, after all, not a video server. Because of that tricky category name, you may have just lost a sale.

So let’s start this discussion by examining the SSR212MC2 as a case in point and see where the storage server concept succeeds and fails.

The Star of the Server. Without the right motherboard, you don’t have an optimal server solution. Intel’s S5000PSLSAS packs in a dizzying number of management-oriented value-adds for a wide range of apps.

Hardware Highlights
“Over the years, storage has been dominated by the big iron guys: EMC, HP, IBM,” says Intel’s Frank Poole, senior technical marketing engineer, server storage products. “It wasn’t until companies like Intel and Supermicro could make servers more of a commodity that it became affordable to do storage servers without having to go to a big iron guy. Our job is to show that it’s really not a big secret. If you look under the hood of an EMC box, for example, you’ll find a server board. Yes, it’s their custom server board, but it’s just a server board with a couple of Intel processors on it, a bunch of hard disks up front, and maybe some software RAID. They’re running a Windows OS on some of their boxes. So for the SMB or small to medium enterprise, we’re just trying to make things more affordable for them so they can retain more storage.”

Pop the hood on an SSR212MC2 and you’ll find that the key components are the S5000PSLSAS (“Star Lake”) motherboard, two 5000-series Xeon sockets, 850W redundant-ready power supply, and (in the SSR212MC2R version) Intel’s SRCSAS144e (“Boiler Bay”) SAS RAID controller card. None of these things by themselves are particularly exceptional in the server world, but put them all together in a management-oriented 2U chassis and now you’ve got something special. Let’s break this down a bit.

Since this barebones server is based on a 5000-series chipset motherboard, you can install a pair of 53xx Xeon CPUs, totaling eight processor cores. Because of thermal constraints, many 2U storage systems use lower-end chips, such as the Celeron. Intel is able to pull out the big guns here owing to the considerable engineering that went into the server’s cooling. A series of 10 hot-swappable fans pull air from the front of the system, past the 12 front-mounted hard drives, and through the vented midplane board. A sculpted plastic shroud channels air past the all-copper, low-profile CPU heatsinks, over the eight FB-DIMM memory slots, and out through a backplane exhaust grille. Traditionally, you would only expect to have six drives in a dual-socket, 2U server, and it’s a testament to Intel’s capabilities that buyers can fit 12 SAS or SATA drives into the low-profile form factor without compromising on performance.

The 12 drives each plug into a SAS/SATA connector on the midplane. In turn, the midplane features a SAS expander module that channels all of the disk traffic to an add-in SAS or SATA RAID controller. The SSR212MC2R version of this server includes Intel’s SRCSAS144e, one of the few top-notch, half-height storage controllers on the market. Fueled by an LSI RAID chip, this card supplies 12 Gbps (four aggregated 3 Gbps lanes) of bandwidth out to the midplane’s expander over its single data connector. When you consider that the fastest 15,000 RPM SAS drives average about 120 MBps, or just shy of 1 Gbps, you can see that Intel’s card is a perfect fit for a 12-drive unit.

Intel takes other steps with this server to ensure the best possible disk performance. Rather than burden the primary storage drives with operating system or application file accesses, the S5000PSLSAS motherboard integrates a discrete LSI 1064e SAS controller capable of running up to two 2.5” inch SAS or SATA hard drives. These secondary drives mount in their own cage located behind the hot-swap fans and can be configured as either RAID 0 or RAID 1. Alternatively, the system’s parallel ATA port can host a fl ash-based Disk On Module (DOM), as can an on-motherboard USB port, USB pin header, or any of the external USB ports on the chassis’ front or back panels.

Of course, there’s a lot more to a “storage server” than the hard drive subsystem. The means by which outside users can access the data on those drives is no less important. Intel integrates a pair of its own 82536EB controllers for dual Gigabit Ethernet connectivity (the company will also soon qualify a 10 Gigabit add-in NIC option), but that leaves the problem of processing overhead. The more TCP/IP packets you can pump in and out of a system, the more processing must be done. Estimates span a wide range, but it’s safe to say that a saturated Gigabit Ethernet connection can consume anywhere from 1 GHz to 2 GHz of CPU horsepower when TCP/IP packets are being handled in software. As you can imagine, this level of overhead would slaughter a server with low-performance CPUs. The situation only grows more critical for companies who adopt iSCSI as their storage area network (SAN) technology of choice. iSCSI hinges on passing SCSI protocol over conventional LANs rather than higher cost infrastructures, such as Fibre Channel. For SMBs wanting to aggregate and centralize multiple storage pools, iSCSI is emerging as the most flexible, cost-effective solution on today’s market, but proper iSCSI performance hinges on preventing TCP/IP bottlenecks.

Companies like Alacritech and others have devised TCP Offload Engine (TOE) solutions to help with packet processing in hardware rather than software, thus taking a substantial burden off of the CPU. However, these cards start around $500, and most cost $1,000 or more. Intel sidesteps this expense through its I/O Acceleration Technology (I/OAT), a collection of processing engines located in the CPU(s), chipset, network controller, and firmware on all Intel 5000-series chipset servers. Intel estimates that I/OAT versus software-based packet processing can yield up to 100% improvement in data throughput along with over 30% faster data exchange between system hardware and applications. Moreover, systems leveraging I/OAT should realize up to 40% less CPU utilization.

If you’ve got 25 people all hitting the same server at the same time, that server better be ready to handle the traffic load. This is doubly true of a storage server, because users are more likely looking for larger fi les with longer transfer times than smaller data tidbits. I/OAT is a good example of how the answer to a problem isn’t always to throw more hardware at it. This also goes for Xeon 5300 processors. If the processing required to fuel this or that iSCSI or storage solution, even after TCP offload, amounts to eight performance-oriented CPU cores, then you’re definitely adding value by selling a single SSR212MC2 box rather than building two or three separate server systems.

Slick and Slim. Those needing a 1U form factor should check out Supermicro’s SS6015X-TV. While the server houses only three hard drives, it keeps the 5000-series advantages found in Intel’s SSR212MC2.

Building the Right Box
Even though you’re supplying a high-efficiency storage server, growing companies will inevitably need more than one machine. Once a company passes the two or three server mark, it’s time to investigate load balancing. An unwise reseller will encourage his customer to toss more servers at a bandwidth problem. The smarter, longer-term play is to perform a site analysis on how existing resources, including storage, are being used. Perhaps what’s needed isn’t more servers so much as an expansion of capacity, rebalancing of users to different boxes, or both.

Another thing that increasingly happens as customers need more storage is that, in order to maintain performance efficiency and reliability, storage becomes centralized and segregated from other servers. When this tendency scales enough, you start getting into clustered storage, advanced NAS, and SAN implementations. Again, the SSR212MC2 fits as a solution here—but only if you stop thinking about it as a “storage server” and starting thinking in external storage terms.

For the moment, forget about the idea of servers being used for applications, such as Web serving, Exchange, or VoIP. Instead, let’s consider fi le storage. For our purposes here, there are two types of storage: transactional and nearline. Transactional fi les are datasets being accessed by relatively large numbers of users on a 24x7 basis. Some common examples are things like airline flight data and weather forecasts—datasets with a limited shelf life that eventually retire to nearline status when they will only be needed occasionally for historical reference. Enterprise SATA is quickly emerging as the preferred drive type for nearline storage while the considerably faster SAS is the de facto choice for transactional situations. A strong storage box should support both drive types through hardware-based RAID.

When transactional speed matters, a storage box needs the flexibility to be able to connect to additional storage resources without sacrificing performance. Two aggregated Gigabit Ethernet connections may be sufficient, but a 1000 Mbps quad-port card or even a 10 Gigabit Ethernet adapter may be necessary. (Note that Intel has multiple SKUs of both.) Fibre Channel, Ultra320 SCSI, and other formats are also options. An external SAS connection is another high-performance option for attaching extra capacity to your server. While it seems trivial up front, the last thing you want is to get caught after the sale realizing that your storage system doesn’t have enough or the correct type of slots for your connectivity. In the case of 2U systems like the SSR212MC2 that use low-profile backplanes, also remember to check card height.

10 Gigs Are Better Than One. For customers needing high-bandwidth, sustained traffic, Intel’s PRO/10GbE CX4 Server Adapter coupled with TCP offloading like I/OAT should serve well.

So to drastically oversimplify this issue, there are basically three types of storage boxes—JBOD, RBOD, and storage server—and which fits a given environment will depend largely on the amount of load placed on that box. A JBOD (Just a Bunch of Disks) is more or less an empty shell waiting for drives and a connection back to a host system. A JBOD such as Adaptec’s 2U SANbloc S50 uses a pair of 4-lane SAS interconnects. An RBOD is just like a JBOD except that it adds hardware RAID acceleration into the enclosure. The storage server is an RBOD with CPU and memory processing resources. The more transactional the data, or the more processing that must be done on that data as it passes in or out of the enclosure, the more likely it is that a customer will benefit from adding compute resources to his storage box.

As an example, consider a magazine’s published content. Articles published a year ago may still be read on occasion by those doing background research, but the access load will be low. A low-cost JBOD will be fine if that data is safely archived somewhere that it can be obtained in case of a system failure. For a magazine’s current content, an RBOD might be more advisable. There will be higher access traffic, response time is more important, and downtime can’t be tolerated, hence the need for a strong RAID configuration. And if that magazine also publishes a wide range of timely, high-resolution video content, then a storage server likely makes the most sense because of all the real-time decoding work that must accompany the many video streams going out. Why not use a standard issue server tower or even a desktop chassis tied to a JBOD or RBOD? Often, the answer comes down to the cost and space consolidation inherent in a unified box like the SSR212MC2.

“We’re seeing a lot of interest in this 2U, 12-drive form factor in the IPTV space, the video-on-demand space,” says Intel’s Frank Poole. “Guys that typically used much more expensive systems to do this are saying, ‘Wow, we can do this at a much lower cost!’ Alternatively, we’re also seeing a lot of interest in the high-performance area, where they say, ‘I can throw my server and all my storage in one box, and it really eliminates bottlenecks, because I don’t have the storage attached outside the box, away from my compute power.’ A huge compute node with huge disk capacity? That’s what these guys have been looking for all along.”

Now, consider the flip-side of this discussion: What if a customer actually does need an application server? Application servers are becoming increasingly common for Web-based tasks such as e-commerce and content management, and in some cases they amass a lot of data. Owners always need to anticipate future growth. We don’t need to rehash the many reasons why a business would want the reliability and performance of a server over a desktop platform. Instead, the question is whether it makes sense to put a server motherboard in a general-purpose desktop- or even server-class case.

The only way to answer this question is to consider what you lose with a non-optimized chassis solution. Obviously, you lose floor space if you’re dealing with a tower or pedestal. You may not get features such as tool-less assembly, and even if you do get it, you’re still going to lose time in additional assembly. You may lose an integrated SAS expander, redundant and hot-swappable power supplies, and, in the case of a non-optimized 2U or 1U rack product, some add-in board options. Most 2U systems also only offer six drive bays. Even a refined 2U, dual-CPU, Intel 5000 chipset unit like Supermicro’s 6025B only provides eight hot-swap bays. When 2U systems offer more, it is often at the expense of CPU performance because of thermal considerations.

In short, a system like the SSR212MC2 makes for an excellent server of any type provided that the user wants a lot of storage expansion room and will benefit from having that server in a rack environment. Buyers get the whole enchilada: maximum storage capacity, flexible expansion and connectivity, up to eight CPU cores (and there’s nothing saying the machine has to start off with two CPUs if money is an issue), and a remarkably compact form factor. You could stack 21 of these machines—up to 168 CPU cores and 252TB of storage (with 1TB drives)—in a single 42U rack. And again, if you’d thought of the SSR212MC2 as a “storage server” rather than an “application server,” you would have missed this solution.

Blades or 1U machines can get you more CPU density, and 3U or higher machines can yield greater storage density, but the 2U design exhibited by the SSR212MC2 offers the best compromise of attributes and does so at whitebox price levels that look nothing like “big iron” options. This is why Intel expects the SSR212MC2 to become the industry’s favored rack server form factor. Intel doesn’t care so much which server brand or model people select; the platform inside is what matters.

That said, there are two specific benefits to buying the SSR212MC2(R) over alternatives. The first is Intel’s world-famous server products support. With a three-year warranty, advanced warranty replacement (AWR), and Intel-direct technical support, nobody offers a more compelling support plan. The other is Intel’s system management software. Typically, the motherboard will have a management console for monitoring and adjusting processor temperatures, fan speeds, and such. A network card will offer another management UI, you can expect a third from the storage controller, and so on. But with the SSR212MC2R and Intel’s PRO Server network adapters, you get a single, unified management console for the entire system that can alert remote IT staff, make for easier total administration, and reduce the chance of disparate management apps conflicting with one another.

Getting Short on Storage. Intel’s SRCSAS144e is one of the few top-quality, half-height controllers available. Pictured here are the storage card’s battery pack and a 1-to-4 splitter cable.

The Software Side
Speaking of software, we should point out that a storage server is just a pile of metal until you put an operating system on it. Picking the right OS is the tricky part. There are plenty of names to choose from. In the past, Intel has followed the big iron model and pre-bundled storage software with its storage servers. The SSR212MA included software from Lefthand Networks. The SSR212PP was essentially a rebranded EMC box loaded with EMC software.

But with the SSR212MC2, Intel has decided to take the whitebox approach to heart and not bundle anything, thus keeping prices down and application options wide open. Of course, not all storage OSes are created equal, and some will work better than others on any given platform. This is why Intel has worked with 15 ISVs just on the SSR- 212MC2 system and made sure that distributors are able to offer and support persuasive hardware/software bundle deals. Wasabi’s Storage Builder for IP-SAN is certified for the SSR212MC2 and delivers robust iSCSI, RAID, and storage management functionality from a tiny fl ash memory DOM.

Intel doesn’t endorse one storage OS over another. Each product has its own merits that will appeal to different customers and resellers with different areas of expertise. However, for system builders and VARs who might be new to the storage world, or for buyers who have standardized on Microsoft’s offerings, there’s a lot of excitement about bundling the SSR212MC2 with Microsoft’s follow-up to Windows Storage Server, the newly released Windows Unified Data Storage Server (WUDSS) 2003.

“I believe this will become the standard for operating systems in the storage space, especially for SMBs,” says Sonny Banga, storage systems product marketing manager at Intel. “Their first-gen product, Windows Storage Server, was file-based only and somewhat limited in feature set. WUDSS is a pretty good next-gen product; we’re excited about it. We think that WUDSS with McKay Creek makes a really compelling value for a reseller in the SMB space. It’s basically a standard for storage appliances.”

WUDSS is based on the conventional Windows Server foundation, but the system has been streamlined and optimized specifically for network storage. One of the ways in which Microsoft does this is by enabling file and block storage. File storage is typical of NAS devices while SANs utilize block storage. The key difference between the two is that block-level storage can transfer pieces of a file, such as an Exchange database, so that the file doesn’t become unavailable to other LAN users during transfer. This is one of the ways that a flexible storage server can work as either a NAS or SAN appliance. Without the ability to run an onboard operating system, you just have a dumb box.

This storage OS is particularly kind to newbie storage resellers in that it starts off installers with a series of task screens to walk admins through configuration. With this done, all subsequent management can be done remotely through a browser interface, so you don’t need to be a command prompt whiz. WUDSS lets you manage multiple storage boxes through a client using Active Directory, and soon you’ll be able to span volumes across storage pools with just a few mouse clicks. Additionally, WUDSS integrates the Microsoft iSCSI Software Target that helps complete the solution as a SAN device.

“Our iSCSI software makes any external disk a target for storing data from any business application as block data,” notes Microsoft group product manager Bala Kasiviswanathan. “We can also boot from external iSCSI targets, so you can create multiple server boot operating systems off this particular target. That’s done by the initiator software. We also include what we call multipathing, which allows people to actually do very robust redundancy in the network. You can easily do failover, load balancing when network traffic is heavy. All that is built into our iSCSI software initiator, and the target software is basically for your provisioning of storage in a bunch of disks at the back end. All of this is now available and affordable to SMB customers, and now it’s possible for a channel partner to build these solutions.”

Give Quad the Nod. Many servers are slot-constrained but hungry for bandwidth. For fiber optic networks, Intel’s PRO/1000 PF Quad Port Server Adapter delivers four GbE ports in one PCIe slot.

The Hot Play
Software systems like WUDSS have brought most of the functionality found in high-end, enterprise storage systems down to a fraction of the enterprise price. For quickly scaling SMBs, a strong storage server plus a smart software package can equal a very new, compelling opportunity for resellers. Expertise in storage obviously helps, but it’s no longer a barrier to entry in this sub-market.

System builders and VARs need more product categories in which they can offer value above and beyond what tier-one competitors can provide. The SSR212MC2 is one of the best tools on today’s field for building niche strength and differentiation. This may be the key to offering a smarter server than you’ve been able to do in the past; it may also unlock success in the storage market. Get familiar with this server, learn its strengths and limitations, then experiment with various OS options. Soon enough, you’ll be racking up sales and higher margins in a space where few SMB players have so far ventured.