[lustre-discuss] LNET Routing Question

Michael Di Domenico mdidomenico4 at gmail.com
Wed May 9 09:26:47 PDT 2018


On Wed, May 9, 2018 at 12:07 PM, Makia Minich
<makia at systemfabricworks.com> wrote:
>> On May 9, 2018, at 10:02 AM, Michael Di Domenico <mdidomenico4 at gmail.com> wrote:
>>
>> On Wed, May 9, 2018 at 9:50 AM, Makia Minich
>> <makia at systemfabricworks.com> wrote:
>>>
>>> I have an LNET routing question. I’ve attached a quick diagram of the current setup; but basically I have two core networks (one infiniband and one ethernet) with a set of LNET routers in between. There is storage and clients on both sides of these routers and all clients need to see all/most storage. All connections, configurations, etc are all working.
>>>
>>> The question is, if an LNET router goes down (which does cause some amount of reconnect or remapping for any clients attempting to use those routes) would this cause any issues or delays for a client’s connection to non-routed storage? Put slightly different, if a job on the ethernet clients is actively using ethernet storage and the lnet routers go down, will job be affected? What about a new job just launching when that lnet router is down?
>>
>> just for the sake of clarity when you say "routers down" do you mean
>> all routers or just one/two?
>
> Thanks for the question, I should have made that clearer. For this question, I was thinking a single router (and no fine-grained routing). I’d also question what would happen if all routers are down: understood that you’d see hangs for any mounts that are LNET-router based, but local-network mounts “should” remain unaffected, right?


unfortunately i can't lend any advice, my theory though would be the
same as yours.  losing a single or multiple routers just cuts the
bandwidth between the client and the storage.  and losing all the
routers would stop connectivity to the remote storage from the
respective client.  i would not expect that loss to prevent
communication to other storage servers


More information about the lustre-discuss mailing list