Ubiquitous Internet access has made "always-on" applications much more common. It's not just hospitals and currency trading desks that demand nonstop computing anymore -- it's now a requirement across many companies. Microsoft Corp.'s, well, monopoly position in the operating system market has made Win 2000 a likely platform for such applications. But how fit is Win 2000 to handle such demands? From what I've seen, chief information officers had better look hard before giving the Microsoft software that job.

Bold advertisements about Win 2000 Professional's "crash-resistant" nature would indicate that Microsoft believes its new software is up to the task. The company declares unequivocally that test results show the operating system is 13 times more reliable than Windows 98. Windows 98!

Ouch. That is like saying a new defense system is many times more secure than the Maginot Line. Or having Al Gore say he's 13 times more faithful than his current boss.

Microsoft's claims notwithstanding, initial research into Win 2000's network capabilities by Tolly Research (a sister organization) seems to indicate that whatever inherent basic reliability exists, it drops significantly if you actually want to use Win 2000 to communicate with another device.

Tolly Research is conducting an ongoing study of Win 2000 networking. While not trying to ferret out bugs, the researchers have nonetheless uncovered bugs aplenty. The bugs are of minor importance; once identified, they can be fixed. More important is the general sense one gets of Microsoft's direction in key areas associated with "reliability."

Let's look at a few examples:

Avoiding dead gateways. While the core of the Internet is robust, the edge is not. Traditionally, each endstation knows about just one "gateway" to the rest of the IP world. If that relay point becomes unavailable, outside communication ceases.

Network infrastructure vendors are on the case, though. In the past few years, through the Internet Engineering Task Force, it has developed and implemented the Virtual Router Redundancy Protocol (VRRP). This protocol adds significant resilience by letting another router port take over the identity of a failed port. We tested this and it works.

Microsoft has decided to solve the problem on its own by having Win 2000 detect the outage and choose an alternate gateway. Network architects might be lulled into thinking this obviates the need for VRRP.

Such is not the case. Many months of research show the Microsoft solution to be poorly conceived, documented and implemented. Sometimes it works, sometimes it doesn't -- hardly ideal for "always-on" applications.

Strengthening the server link. Link aggregation is an industry-standard method for having two or more server network interface cards (NIC) behave as one. In addition to the bandwidth benefits, this approach removes the NIC, cabling and switch ports as a single point of failure.

Tolly Research studies show hardware vendor support for this function to be weaker under Win 2000 than under NT 4. Worse, using the link aggregation of a major NIC vendor, engineers can get Win 2000 server to "blue screen" like clockwork. With another, failover works but overall bandwidth delivery is weak.

When implementing Win 2000 reliability features actually reduces reliability, it is clear that a much closer look at this software is in order.

» posted by abennett

Network World

• Email
• Print
• Share
  • Slashdot
  • Digg
  • Stumble
  • Reddit
  • Newsvine