[lustre-discuss] JBOD or Linux Software RAID0 for OSTs with HA
Pinkesh Valdria
pinkesh.valdria at oracle.com
Wed Jan 8 11:04:10 PST 2020
I wanted to ask, what is the best practice when it comes to building the OST. We have Block volumes we can provision and each of them can deliver 480 MB/s throughput and they can be attached to multiple compute nodes in sharable read/write mode. I have a 25Gbps (3125 MB/s) network between OSS and Storage devices (aka: block volumes). Lustre clients nodes also have 25Gbps network connection to OSS nodes.
Options:
On OSS servers, I setup 7 or more OST’s (7*480=3360MB/s) and set Stripe Count to 7 or more, to ensure clients get max throughput performance possible.
On OSS servers, I setup a Linux software RAID0 of 7 block volumes, set Stripe count = 1 , to ensure clients get max throughput performance possible. I can setup multiple OSTs for the OSS, with each OST been a RAID0 of 7 block volumes.
Question: Is it better to use Option2 or let clients use stripe count 7 or more, to get the max performance from the file system. Most clients will use 1 File per process when writing files to file system. With Option2, I am trying to hide the Storage backend throughput complexity from clients.
Our Block volumes are already replicated, so I don’t need to do RAID1, RAID5 or RAID6 at the Linux s/w RAID level. In option2, I am using RAID0, so I can get throughput from 7 Block volumes which matches the network speed I will have between most of my Lustre clients and Lustre OSS servers.
Another question is, has anyone implemented HA across 2 OSS nodes pair/building block and using OST’s which are linux software RAID0. Issue I am facing is after I create RAID0 on OSS1, I have to reboot OSS2 to make it visible on OSS2. Then if I reboot OSS1, the RAID0 (/dev/md127) is gone on OSS1. Wondering if this is technically even feasible to have HA with Linux software RAID0 ?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20200108/c3a1de6c/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 179622 bytes
Desc: not available
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20200108/c3a1de6c/attachment-0001.png>
More information about the lustre-discuss
mailing list