[LU-10300] Can the Lustre 2.10.x clients support 64K kernel page?

• Type: Question/Request
• Resolution: Cannot Reproduce
• Priority: Major
• Affects Version/s: Lustre 2.10.0
• Environment: Redhat
• Assignee: James A Simmons
• Reporter: ZhangWei (Inactive)

Description

We test Lustre in the kernel 4.11.0 which use 64K memory page, and we found that there are no config items in the configure.

Issue Links

• is related to: LU-12255 sanity-dom test 42e fails with ‘test_42e failed with 1’
• is related to: LU-4398 mdt_object_open_lock() may not flush conflicting handles
• is related to: LU-10073 lnet-selftest test_smoke: lst Error found
• is related to: LU-11729 ARM: sanity test_810: BAD WRITE CHECKSUM with adler
• is related to: LU-11595 sanity-dom sanityn test 11: LBUG: (file.c:470:ll_dom_finish_open()) ASSERTION( lnb.lnb_file_offset % ((1UL) << 16) == 0 ) failed
• is related to: LU-11596 sanity test_42d/test_42e: FAIL: failed: client:143294464 server: 143110144
• is related to: LU-12014 check correct size in ll_dom_finish_open()
• is related to: LU-15722 IO write gets stuck on some sanityn test cases for 64K PAGE_SIZE
• is related to: LU-11667 sanity test 317: FAIL: Expected Block 8 got 48 for f317.sanity
• is related to: LU-11671 sanity test 45: FAIL: write wasn't cached
• is related to: LU-11787 sanityn test 71a fails with ‘data is not flushed from client’
• is related to: LU-12362 kernel warning 'do not call blocking ops when !TASK_RUNNING ' in ptlrpcd
• is related to: LU-12419 ppc64le: "LNetError: RDMA has too many fragments for peer_ni" when reading two files
• is related to: LU-15223 Improve partial page read/write
• is related to: LU-14970 sanity-flr test_50a: FAIL: Mirrored file consumes 1728 blocks
• is related to: LU-14346 sanity-pcc test 7a hangs in multiop/down_read
• is related to: LU-10157 LNET_MAX_IOV hard coded to 256
• is related to: LU-15364 Kernel oops when stripe on Arm64 Server end multiple MDTs
• is related to: LU-7650 ko2iblnd map_on_demand can't negotitate when page sizes are different between nodes.
• is related to: LU-11597 sanityn test 16a failed with direct I/O
• is related to: LU-11785 conf-sanity test_98 fails with 'Buffer overflow check failed'
• is related to: LU-11788 sanity test 104a fails with ‘lfs df failed’
• is related to: LU-6387 Add Power8 support to Lustre
• is related to: LU-11200 Centos 8 arm64 server support
• is related to: LU-15293 Add lbuild support for latest Arm64 CentOS8

Comments

[James A Simmons]

So I have managed to get Lustre ZFS servers running on Power8 nodes. I got it to mount and then it just locked up with any attempt to use the file system  Well its a start

[James A Simmons]

Once the RHEL ARM/Power8 server support work is complete we can test 64K pages on the server side.

[Andreas Dilger]

It probably makes sense to print a build warning in the server code when PAGE_SIZE isn't 4096, just so people are aware that this isn't being tested. If someone starts testing this in the future, the warning can be removed.

[ZhangWei (Inactive)]

Yes, I tested the 64KB PAGE_SIZE server with 4KB PAGE_SIZE Client and 64KB PAGE_SIZE Client, there are some errors both in these two clients.Thanks for help !

[Andreas Dilger]

We definitely used to run Lustre with 64KB PAGE_SIZE on IA64 clients, and in theory this would still work but we haven't tested it in a long time.  We never had much success with 64KB PAGE_SIZE on the server, since this caused problems with 4KB PAGE_SIZE clients doing writes to the 64KB PAGE_SIZE server.

[ZhangWei (Inactive)]

Thanks for your replay, with the map_ondemand enabled, can the Lustre only support 4k kernel memory page?

[James A Simmons]

Actually I have tested such a setup on Power8 which is 64K page based. The only issue that came up was when dealing with the ko2iblnd witth map_on_demand enabled. Otherwise it works.