Earlier this month, I wrote about agent and agentless management approaches. That article elicited a few responses from readers, which I shared with you in a subsequent issue. I have just received another response, this time from a reader who sent me his thoughts on Intelligent Platform Management Initiative, a hardware manageability interface specification that embeds management in the hardware.
IPMI is a message-based interface between CPUs and hardware devices of different types. It enables remote out-of-band management, which allows administrators to contact a server even when the operating system is not operational. Many hardware manufacturers have adopted IPMI, and servers are shipping with this embedded capability. So the question is, do you even know that IPMI is on your servers? And if so, are you taking advantage of it?
Here's what our reader, a consultant, had to say about IPMI in his response to the reader-feedback newsletter: "One reader suggested agentless was not good because of network overhead. Generally speaking, my issues are usually not in the network (most of our customers are over provisioned on their links), rather, the overhead of [operating system]-based agents on the server it's installed on. While it might have the smarts to only send relevant information, and so reduce network traffic, it's consuming memory and CPU cycles.
"Actually, most of our customers are using new PowerEdge servers from Dell that come with IPMI. What's really neat with this is that the agent is prebuilt into the motherboard, so it meets my concerns of not stealing memory/CPU cycles (I'm told it runs on a [baseboard management controller] which has its own [real-time operating system]). But the difference is that it also has the smarts to only wake up when it needs to - to send me alerts based on thresholds I set - so reducing network traffic."
He goes on to say, "[Another] reader talked about the lack of data leading up to a server crash if you use agentless - well, here comes IPMI again. It has what's called an event log that captures hardware-based events - like a fan kicking in due to a temperature spike (happens a lot in some of our customers' closets) - or someone popping the chassis to take a look inside... Best thing is that the event log is in the BMC - so it's always going to be there and is a great way to show the server failure isn't at the application level."
The reader continues: "Admittedly, IPMI is all well and good for hardware. You invariably also need agents to track application-level issues. So the two worlds are best used together. And given IPMI is free with Dell servers, there's really no excuse not to use it."
As a parting note, this particular reader mentioned Dell servers, but IPMI also ships on a variety of servers delivered by other hardware manufacturers. So the key here is not necessarily the brand, it's the IPMI functionality. So if you weren't aware of, or are not using IPMI, check to see if it is on your newer servers. And if it is, take advantage of it.
Learn more about this topicAgent or agentless monitoring? It's your choice
Network World, 06/08/05Mailbag: Agents and agentless management
Network World, 06/15/05CA launches security pack for SMB
Network World, 06/27/05