On Tuesday, 02/17/2015 at 05:33 EST, Berthold Gunreben <b...@suse.de> wrote: > Well, I don't see it as an advantage over real link aggregation, but > IMHO it is a real advantage over an active/passive failover.
It's worse. Keep those packets in order! > You might argue that LACP is available to z/VM, but the downside is, > that it subverses the virtualization of the OSA adapters, because it > reserves the port. > > Or is there a real link aggregation available in z/VM that works > without a port group? Means, is there s link aggregation in z/VM that > works on shared OSA adapters? >From the z13 announcement: z/VM Multi-VSwitch Link Aggregation Support With the PTFs for APARs VM65583 and PI21053, z/VM V6.3 provides Multi-VSwitch Link Aggregation Support, allowing a port group of OSA-Express features to span multiple virtual switches within a single z/VM system or between multiple z/VM systems. Sharing a Link Aggregation Port Group (LAG) with multiple virtual switches increases optimization and utilization of the OSA-Express when handling larger traffic loads. Higher adapter utilization protects customer investments, which is increasingly important as 10 gigabit deployments become more prevalent. This enhancement makes it possible to do VSwitch Link Aggregation with OSAs shared with other z/VM logical partitions, lifting the previous restriction of requiring dedicated OSAs. > The advantage that I see here for Linux/KVM is, that you can run > balance-rr on shared OSA devices. And while there is probably not much > of a difference to the Mainframe side, the Network Switches are way > more happy with this mode than with active/passive. The network switch is neither happy nor sad. :-) As long as you don't have excessive queuing or service times for the adapter, you don't need more than one. And unlike x86 devices, an OSA runs more "smoothly" when you keep data running through it. Start/Stop/Start/Stop just wastes time. > > And one related cautionary note: I recently learned that you > > shouldn't put unbonded adapters on the same subnet without doing your > > homework. By default, Linux will consolidate all traffic onto just > > one of those adapters, it having all of the same-subnet IP addresses > > consolidated onto it. (Very odd when you see QUERY VSWITCH DETAILS > > output showing adapters with no IP addresses and others with more > > than one!) I'm not really sure *why* Linux does that, but I'm > > guessing that it's an x86 thing. Or, it could just be to annoy me. > > Is it really linux that does this? I normally see, that without a port > group, you have an active adapter, and a backup adapter for a VSWITCH. Yes, it's really true. If your routing table and the return path filter (rp_filter) aren't in agreement, packets get discarded. Linux moves the IP addresses to a single adapter. See http://www.novell.com/support/kb/doc.php?id=7007649 Alan Altmark Senior Managing z/VM and Linux Consultant Lab Services System z Delivery Practice IBM Systems & Technology Group ibm.com/systems/services/labservices office: 607.429.3323 mobile; 607.321.7556 alan_altm...@us.ibm.com IBM Endicott ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/