Oracle wants to improve Linux load balancing and failover

Oracle reckons Linux remote direct memory access (RDMA) implementations need features like high availability and load balancing, and hopes to sling code into the kernel to do exactly that. The problem, as Oracle Linux kernel developer Sudhakar Dindukurti explained in this post, is that performance and security considerations …

[Anonymous Coward]

Wish I could upvote an article... trying to read this at 8:45am, whilst still half a sleep, left me going.. huh? what? huh? umm?? It did eventually make sense though :D

[Multivac]

Stop fiddling with other peoples stuff

There are many different scenarios that may cause you to have to fail over, this addresses just one of those scenarios. If Oracle could maybe put a little more effort into updating their applications to be more HA then we’d likely get a heck of a lot more of those scenarios covered.

[Anonymous Coward]

Re: Stop fiddling with other peoples stuff

Isn't fiddling with other people stuff the very reason of open source? Or do you believe it's only software you don't have to pay for?

It's up to kernel maintainers then to accept Oracle's code or not.

[Gerhard Mack]

Re: Stop fiddling with other peoples stuff

They never consider or test anyone else' scenario. I ran into that the hard way with OCFS. It was designed and tested for DB workloads but freakishly unstable for anything else.

[Anonymous Coward]

Re: Stop fiddling with other peoples stuff

When it's open source is up to you to extend and add the missing pieces, isn't it? Of course Oracle is interested more in its database business than generic workloads.

[Gerhard Mack]

Re: Stop fiddling with other peoples stuff

I'm not talking bad performance, I'm talking about crashes involving data loss.

[LeoP]

Fair and square

It is exceptional, that Oracle does (or tries to do) something good. Let's take a minute to appreciate this - god knows, when this will happen again.

[hammarbtyp]

Re: Fair and square

I know...suspicious isn't it

[Jim Preis]

Wasn't HPs Moonshot kinda-sorta about accessing gobs and gobs of memory-speed storage using photonics as the medium? The reason I'm asking is because despite being a registered member of Team Lenovo, I recall being really inspired by HP's whole Moonshot architecture which, had it been built, would have rendered RDMA less important. It seems like we'll get to something like Moonshot, but incrementally and likely over decades.

[johnnybee]

You're thinking of 'The Machine'

Moonshot is a box of nodes with an internal switched network....

[Peter Gathercole]

IBMs RDMA over HFI and IB

ran RDMA over an abstracted network device, with the resilience built in to the underlying network. This allowed the network layer to adjust to failures without the RDMA setup being exposed to the changes.

Seemed to work quite well on AIX, and I believe that the Linux support (for the P7 775 9125-F2C) worked the same (or even better!). I'm pretty sure that IBM would have put their work back into the kernel.

[Anonymous Coward]

Nooooooo!

Don't let Crazy Larry near it, he'll kill Tux and turn him into a rug for the yacht!

[Anonymous Coward]

Linux remote direct memory access (RDMA)

What mitigations did the designers build in to prevent RDMA being used in security exploits.

[Anonymous Coward]

Transparent interface bonding and failover has been in the mainline Mellanox IB adapter implementation for years.

Not really seeing what this is supposed to be adding?