<div dir="ltr"><div>Hi Erik,</div><div><br></div><div>It's great to hear positive feedback! Thanks for taking out time to send out this email. It means a lot to us :) <br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Thu, Apr 9, 2020 at 10:55 AM Strahil Nikolov <<a href="mailto:hunter86_bg@yahoo.com">hunter86_bg@yahoo.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">On April 8, 2020 10:15:27 PM GMT+03:00, Erik Jacobson <<a href="mailto:erik.jacobson@hpe.com" target="_blank">erik.jacobson@hpe.com</a>> wrote:<br>
>I wanted to share some positive news with the group here.<br>
><br>
>Summary: Using sharding and squashfs image files instead of expanded<br>
>directory trees for RO NFS OS images have led to impressive boot times<br>
>of<br>
>2k diskless node clusters using 12 servers for gluster+tftp+etc+etc.<br>
><br>
>Details:<br>
><br>
>As you may have seen in some of my other posts, we have been using<br>
>gluster to boot giant clusters, some of which are in the top500 list of<br>
>HPC resources. The compute nodes are diskless.<br>
><br>
>Up until now, we have done this by pushing an operating system from our<br>
>head node to the storage cluster, which is made up of one or more<br>
>3-server/(3-brick) subvolumes in a distributed/replicate configuration.<br>
>The servers are also PXE-boot and tftboot servers and also serve the<br>
>"miniroot" (basically a fat initrd with a cluster manager toolchain).<br>
>We also locate other management functions there unrelated to boot and<br>
>root.<br>
><br>
>This copy of the operating system is a simple a directory tree<br>
>representing the whole operating system image. You could 'chroot' in to<br>
>it, for example.<br>
><br>
>So this operating system is a read-only NFS mount point used as a base<br>
>by all compute nodes to use as their root filesystem.<br>
><br>
>This has been working well, getting us boot times (not including BIOS<br>
>startup) of between 10 and 15 minutes for a 2,000 node cluster.<br>
>Typically a<br>
>cluster like this would have 12 gluster/nfs servers in 3 subvolumes. On<br>
>simple<br>
>RHEL8 images without much customization, I tend to get 10 minutes.<br>
><br>
>We have observed some slow-downs with certain job launch work loads for<br>
>customers who have very metadata intensive job launch. The metadata<br>
>load<br>
>of such an operation is very intensive, with giant loads being observed<br>
>on the gluster servers.<br>
><br>
>We recently started supporting RW NFS as opposed to TMPFS for this<br>
>solution for the writable components of root. Our customers tend to<br>
>prefer<br>
>to keep every byte of memory for jobs. We came up with a solution of<br>
>hosting<br>
>the RW NFS sparse files with XFS filesystems on top from a writable<br>
>area in<br>
>gluster for NFS. This makes the RW NFS solution very fast because it<br>
>reduces<br>
>RW NFS metadata per-node. Boot times didn't go up significantly (but<br>
>our<br>
>first attempt with just using a directory tree was a slow disaster,<br>
>hitting<br>
>the worse-case lots of small file write + lots of metadata work load).<br>
>So we<br>
>solved that problem with XFS FS images on RW NFS.<br>
><br>
>Building on that idea, we have in our development branch, a version of<br>
>the<br>
>solution that changes the RO NFS image to a squashfs file on a sharding<br>
>volume. That is, instead of each operating system being many thousands<br>
>of files and being (slowly) synced to the gluser servers, the head node<br>
>makes a squashfs file out of the image and pushes that. Then all the<br>
>compute nodes mount the squashfs image from the NFS mount.<br>
> (mount RO NFS mount, loop-mount squashfs image).<br>
><br>
>On a 2,000 node cluster I had access to for a time, our prototype got<br>
>us<br>
>boot times of 5 minutes -- including RO NFS with squashfs and the RW<br>
>NFS<br>
>for writable areas like /etc, /var, etc (on an XFS image file).<br>
> * We also tried RW NFS with OVERLAY and no problem there<br>
><br>
>I expect, for people who prefer the squashfs non-expanded format, we<br>
>can<br>
>reduce the leader per compute density.<br>
><br>
>Now, not all customers will want squashfs. Some want to be able to edit<br>
>a file and see it instantly on all nodes. However, customers looking<br>
>for<br>
>fast boot times or who are suffering slowness on metadata intensive<br>
>job launch work loads, will have a new fast option.<br>
><br>
>Therefore, it's very important we still solve the bug we're working on<br>
>in another thread. But I wanted to share something positive.<br>
><br>
>So now I've said something positive instead of only asking for help :)<br>
>:)<br>
><br>
>Erik<br>
>________<br>
><br>
><br>
><br>
>Community Meeting Calendar:<br>
><br>
>Schedule -<br>
>Every 2nd and 4th Tuesday at 14:30 IST / 09:00 UTC<br>
>Bridge: <a href="https://bluejeans.com/441850968" rel="noreferrer" target="_blank">https://bluejeans.com/441850968</a><br>
><br>
>Gluster-users mailing list<br>
><a href="mailto:Gluster-users@gluster.org" target="_blank">Gluster-users@gluster.org</a><br>
><a href="https://lists.gluster.org/mailman/listinfo/gluster-users" rel="noreferrer" target="_blank">https://lists.gluster.org/mailman/listinfo/gluster-users</a><br>
<br>
Good Job Erik!<br>
<br>
Best Regards,<br>
Strahil Nikolov<br>
________<br>
<br>
<br>
<br>
Community Meeting Calendar:<br>
<br>
Schedule -<br>
Every 2nd and 4th Tuesday at 14:30 IST / 09:00 UTC<br>
Bridge: <a href="https://bluejeans.com/441850968" rel="noreferrer" target="_blank">https://bluejeans.com/441850968</a><br>
<br>
Gluster-users mailing list<br>
<a href="mailto:Gluster-users@gluster.org" target="_blank">Gluster-users@gluster.org</a><br>
<a href="https://lists.gluster.org/mailman/listinfo/gluster-users" rel="noreferrer" target="_blank">https://lists.gluster.org/mailman/listinfo/gluster-users</a><br>
<br>
</blockquote></div><br clear="all"><br>-- <br><div dir="ltr" class="gmail_signature">Regards,<br>Hari Gowtham.</div>