When was the last time you swapped out the processors on a production server? Have you ever ripped out a working system’s RAID controller and substituted one with bigger cache? How about pulling out a machine’s mirrored 18GB Ultra160 SCSI boot drives just to replace them with some 36GB Ultra360 spindles?
Despite the fact that top-tier server manufacturers boast about the field upgrade capabilities of their server platforms, it’s a myth that anyone ever fiddles with a production system except to replace a blown part. If the server is less than a year old, chances are that it was ordered with the right parts and doesn’t need to be touched. If the server is more than year old, nobody in their right mind is going to pop the top to crank the gigahertz.
To research this myth, I contacted all the tier-one server manufactures. Not one would formally cooperate when asked for statistics regarding enhancements to their servers, either by sales of upgrade parts or through calls made by their field-service teams. Some said the data wasn’t available. Others said it was proprietary information that couldn’t be released for competitive reasons. All claimed to find the question surprising -- and were interested in reading the results.
Fortunately, one vendor, who shall remain nameless, forwarded the informal comments of a marketing manager, whose name was removed from the e-mail. The manager’s thinking echoed my own: “I believe the majority of customers purchase initially a server populated with the RAM and processors for future growth.”
The manager added, “Many customers secure capital expenditures for the hardware and it is easier to purchase under this capital than to try to expense some more hardware down the line.”
Another reason, of course, for not upgrading a system would include a fear of screwing things up, either by having hardware problems or by encountering difficulties with the operating system, drivers, or applications. Given that there’s going to be only a minimal performance improvement in going from, say, 2.0GHz to 2.6GHz processors while the rest of the server remains the same, what’s the point in taking that risk?
If one could generalize, then one would say: The smaller the server, the less likely its hardware is going to be touched after the system has been deployed. The chassis investment in an eight- or 16-way server might warrant enhancements to its I/O backplane. It also might make sense to add processors, if some of the sockets were initially unpopulated.
By contrast, it’s hard to imagine anyone doing much to the hardware on a dual-processor 1U or 2U server or to a server blade, other than adding memory if needed. If that low-profile server can’t handle the workload, the solution would be to replace it with a more powerful server or to add more servers to a load-balanced cluster. What about swapping the processors or adding a faster backplane? There’s no ROI for spending good money on old servers. When you’re considering specifications for new servers, make sure the system fits your existing needs, and buy it with the headroom you anticipate requiring for the expected life span of the machine. Unless you have an IT culture that actually performs server upgrades, don’t plan on performing any, and don’t pay extra for features such as upgradable CPU cards capable of accommodating future processor platforms. You won’t use them.