Time for another piece of my little series! This time I'd like to write about a new feature in v7.0x especially for administrators and support personnel: The Frame Log. Maybe it's a bit early to write about it, because it seems to be a feature "in development" at the moment, but I did wait for it so long I'm just not able to resist. I think and I hope Brocade will further develop it like the bottleneckmon - which I was very sceptical about in its first version when it was released in the v6.3 code. After seeing its functionality being extended on v6.4 and even more in v7.0, the bottleneckmon is an absolute must-have.
Hmm... maybe I should write an article about bottleneckmon, too )
Basically it is a list of frame discards. There are several reasons why a switch would have to drop a frame instead of delivering it to the destination device. One of them is a timeout. If a frame sticks in the ASIC (the "brain" behind the port) for half a second, the switch has to assume that something's going wrong and so the frame cannot be delivered in time anymore. Then it drops it. Till FabOS v7.0 it just increased a counter by one. Since later v6.2x versions it was at least logged against the TX port (the direction towards the reason for the drop) - in earlier versions the counter increased only for the origin port, which made no sense at all. But now we even have a log for it! A log to store all the frames the switch had to discard. While that sounds a bit like rummaging through the switch's trash bin, the Frame Log is very useful for troubleshooting though. It contains the exact time, the TX and the RX port (keep in mind the TX is the important one) and even information from the frame itself. In the summary view you see the fibrechannel addresses of the source device (SID) and of the destination device (DID).
For example to see the two most recent frame discards in summary mode, just type:
B48P16G:admin> framelog --show -mode summary -n 2 ===================================================================== Fri Sep 23 16:07:13 CET 2011 ====================================================================== Log TX RX timestamp port port SID DID SFID DFID Type Count ====================================================================== Sep 29 16:02:08 7 5 0x040500 0x013300 1 1 timeout 1 Sep 29 16:04:51 7 1 0x030900 0x013000 1 1 timeout 1
In the so called "dump mode" you even see the first 64 bytes of each frame. Usually I have to bring an XGIG tracer onsite to catch such information and often it's not even possible to catch it then, because an XGIG can only trace what's going through the fibre. So you'll only see this frame if you trace a link it crosses before it is dropped. And even then you can't trigger (=stop) the tracer directly on this event, but you have to have it looking for a so called ABTS (abort sequence). If a frame is dropped the command will time out in the initiator and it will send this ABTS. Depending on what frame exactly was dropped in what direction, the ABTS could be on the link several minutes after the actual drop of the frame. Imagine a READ command being dropped. The error recovery will start after the SCSI timeout which could be e.g. 2 minutes. But 2 minutes is a long time in a FC trace. Chances are good that the tracer misses it then.
The frame log can tell you exactly which frame was dropped. If you try to find out if a particular I/O timeout in your host was caused by a timeout discard in the fabric, this is your way to go. If you see your storage array complaining about aborts for certain sequences, just look them up in the Frame Log. With this feature Brocade finally catches up with Cisco and their internal tracing capabilities - and Brocade does it way more comfortable for the admin. The logging of discarded frames is enabled by default and it works on all 8G and 16G platform switches without any additional license.
As I mentioned at the beginning of this article there are still things for Brocade to work on to turn the Frame Log into a must-have tool like the bottleneckmon. The first catch is its volatility. In the current version it can only keep 50 frames per second on an ASIC base for 20 minutes in total. At the moment I personally think that's too short. But I'll wait for the first cases where I can use it before I forge an ultimate opinion about this limit.
The other - more concerning - constraint is that it only works for discards due to timeout at the moment. So if a frame is dropped because of one of all the other possible reasons, it won't be visible in the Frame Log in its current implementation. But that's exactly what I need! If the switch discards a frame because of a zone mismatch or because the destination switch was not reachable or because the target device was temporarily offline or whatever - I want to see that. If a server is misconfigured (uses wrong addresses) and so cannot reach its targets, you'd see the reason right there in the target log - no tracing needed! There are plenty of other situations that would be covered with such a functionality. So I honestly hope that there is a developer with a concept like this in his drawer or even already within its implementation. Allow me to assure you that there is at least one support guy waiting for it...
The picture is from Zsuzsanna Kilian. Thank you!