EC plugin ISA-L production ready?
Hi *, since Tentacle the new default EC plugin is ISA-L, I wanted to ask if anyone is using it in production (maybe for several years already)? I searched tracker for issues with the cauchy technique and found several (haven't read all of them), sometimes in combination with FastEC (which isn't safe to use yet from my understanding), but for example this doesn't mention the technique: https://tracker.ceph.com/issues/74943 and it's not backported to Tentacle yet. So my current impression is that ISA may not be fully production ready, is it? Would it be safer to stick to jerasure for now although it's not being maintained anymore? One specific EC profile I'm dealing with is k=5 m=7 which can be created with "isa + cauchy" or "jerasure + reed_sol_van" but is refused by "isa + reed_sol_van". I'd appreciate any comments! Thanks! Eugen
See https://telemetry-public.ceph.com/d/b5664aff-721e-4c8a-b79a-7d3e2a8eaf07/ec-... which shows the distribution of clusters with telemetry enabled using different EC plugins. Note the log2 Y axis and that not all systems have telemetry enabled. Issue 74943 is a concern if you have multiple EC pools using ISA with different choices of k and m. If all your EC pools use the same values of k and m you don't need to worry about this. There are some big-endian bugs with the version of ISA in Tentacle, if you are running Ceph on exotic hardware (not X86 or ARM) then you'll want to use Umbrella which has a newer version of the library with these fixes. ISA reed_sol_van doesn't support m>4 or m=4 with k>21, this is correctly policed, attempts to create pools with unsupported configurations will fail. Cheers, Bill. bill_scales@uk.ibm.com<mailto:bill_scales@uk.ibm.com> IBM Distinguished Engineer, IBM Storage From: Eugen Block <ceph-users@ceph.io> Date: Thursday, 22 October 2026 at 10:25 To: ceph-users@ceph.io <ceph-users@ceph.io> Subject: [EXTERNAL] [ceph-users] EC plugin ISA-L production ready? Hi *, since Tentacle the new default EC plugin is ISA-L, I wanted to ask if anyone is using it in production (maybe for several years already)? I searched tracker for issues with the cauchy technique and found several (haven't read all of them), sometimes in combination with FastEC (which isn't safe to use yet from my understanding), but for example this doesn't mention the technique: https://tracker.ceph.com/issues/74943 and it's not backported to Tentacle yet. So my current impression is that ISA may not be fully production ready, is it? Would it be safer to stick to jerasure for now although it's not being maintained anymore? One specific EC profile I'm dealing with is k=5 m=7 which can be created with "isa + cauchy" or "jerasure + reed_sol_van" but is refused by "isa + reed_sol_van". I'd appreciate any comments! Thanks! Eugen _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io Unless otherwise stated above: IBM United Kingdom Limited Registered in England and Wales with number 741598 Registered office: Building C, IBM Hursley Office, Hursley Park Road, Winchester, Hampshire SO21 2JN
Thanks, Bill. I saw the telemetry data in a presentation as well, but I'm not sure how many production cluster operators/companies opt-in for telemetry. The scenario I'm currently looking at (for one specific customer) would be using different EC profiles, yes, but no exotic hardware. Since Umbrella isn't even out yet, and the customer wouldn't upgrade a production cluster to a brand new major release, they'll probably have to stick to jerasure for now, it seems. Zitat von Bill Scales <ceph-users@ceph.io>:
See https://telemetry-public.ceph.com/d/b5664aff-721e-4c8a-b79a-7d3e2a8eaf07/ec-... which shows the distribution of clusters with telemetry enabled using different EC plugins. Note the log2 Y axis and that not all systems have telemetry enabled.
Issue 74943 is a concern if you have multiple EC pools using ISA with different choices of k and m. If all your EC pools use the same values of k and m you don't need to worry about this.
There are some big-endian bugs with the version of ISA in Tentacle, if you are running Ceph on exotic hardware (not X86 or ARM) then you'll want to use Umbrella which has a newer version of the library with these fixes.
ISA reed_sol_van doesn't support m>4 or m=4 with k>21, this is correctly policed, attempts to create pools with unsupported configurations will fail.
Cheers,
Bill. bill_scales@uk.ibm.com<mailto:bill_scales@uk.ibm.com> IBM Distinguished Engineer, IBM Storage
From: Eugen Block <ceph-users@ceph.io> Date: Thursday, 22 October 2026 at 10:25 To: ceph-users@ceph.io <ceph-users@ceph.io> Subject: [EXTERNAL] [ceph-users] EC plugin ISA-L production ready?
Hi *,
since Tentacle the new default EC plugin is ISA-L, I wanted to ask if anyone is using it in production (maybe for several years already)? I searched tracker for issues with the cauchy technique and found several (haven't read all of them), sometimes in combination with FastEC (which isn't safe to use yet from my understanding), but for example this doesn't mention the technique:
https://tracker.ceph.com/issues/74943
and it's not backported to Tentacle yet. So my current impression is that ISA may not be fully production ready, is it? Would it be safer to stick to jerasure for now although it's not being maintained anymore?
One specific EC profile I'm dealing with is k=5 m=7 which can be created with "isa + cauchy" or "jerasure + reed_sol_van" but is refused by "isa + reed_sol_van".
I'd appreciate any comments!
Thanks! Eugen
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
Unless otherwise stated above:
IBM United Kingdom Limited Registered in England and Wales with number 741598 Registered office: Building C, IBM Hursley Office, Hursley Park Road, Winchester, Hampshire SO21 2JN _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
participants (2)
-
Bill Scales
-
Eugen Block