According to Techtel, a provider of intelligence in the technology market space, spending plans have dramatically improved over the past months and become very positive, which points to sustainable improvement in the tech sector. While this is not a direct quote and is simply my attempt to decipher the bread crumbs they offer freely, we have to look no farther than our own server farm business to get a sense of the marketplace.  All signs are pointing up and we hope your IT budget is also climbing, so if you are considering replacing any servers or building new ones, this may be an opportune time. 

The industry is recognizing Opteron's outstanding cost-performance ratio.  With the increased memory management and ability to concurrently support 32 and extended 64-bit applications the Opteron is simply a bargain.  For those still building Intel systems it is worth reviewing your decision.  

During the course of this article we are building a new SQL server based on a single Opteron 148 CPU and the ASUS SK8V motherboard.  We are not going into details of the construction which is straightforward but we want an example of a typical server to help illustrate why it is our "bread and butter" machine configuration. In our example server software is actually a significant part of the overall expense. If we were building an Itanium server it could be justified by an application that required the Itanium - but right now it is hard to see exactly what would require the Itanium platform in the "typical" hosting marketplace. 

While the Itanium can outperform the Opteron in floating-point calculations (but not integer calculations) and is a strong entry into the server market, it is comparatively expensive.  The die size is large and cannot be made easily and more importantly unlike the Opteron it was not designed with the ability to use existing software.  This was a costly miscalculation for Intel and their technology partner Hewlett Packard. 

As a hosting company we would rather invest in multiple servers (with some exceptions) that are strong enough for the most demanding database or application running on them.  We divide shared accounts and do not allow one server to handle a large number of accounts.  There are two reasons this.  First, is the approach where you want less eggs in one basket and a more manageable GUI interface to the server applications.   Secondly, it is important to leave a generous amount of unused potential so that you can meet all future needs of the accounts hosted on that particular server, even after they grow in size and demands over a period of years.   We host many of the same domains we did in 1996, the first year of our hosting business.  Some of the more active sites have grown into as many as two or three gigabytes.  Having to move accounts from one server to another to balance the loads is not something that makes you feel like a great planner of resources.  We almost never have to do that because we anticipate and leave ample headroom on our server resources. We know of one proud hosting company that keeps their SQL machines running at 50% CPU load, a guarantee that their customers websites are experiencing some latency and in our view is very unprofessional at the least.