There are both pros and cons to having more PGs. Here are a couple of
considerations:
Pros:
1) Better data distribution prior to balancing (and maybe after)
2) Fewer objects/data per PG
3) Lower per-PG lock contention
Cons:
1) Higher PG log memory usage until you hit the osd target unless you
shorten the per-PG log length.*
2) More work for the mons
3) More work for the mgr per collection interval
* Counter-intuitively increasing the number of PGs in a replicated pool
with simple objects may lower aggregate PG log memory consumption if
there are other EC pools with complex objects being written to
concurrently. This is due to the global limit on PG log entries. See:
osd_target_pg_log_entries_per_osd
Mark
On 3/5/24 14:11, Nikolaos Dandoulakis wrote:
Hi Anthony,
Thank you very much for your input.
It is a mixture of HDDs and a few NVMe drives. The sizes of the HDDs vary between 8-18TB and `ceph osd df` reports 23-25 pgs for the small drives 50-55 for the bigger ones.
Considering that the cluster is working fine, what would be the benefit of more pgs?
Best,
Nick
________________________________
From: Anthony D'Atri <aad@xxxxxxxxxxxxxx>
Sent: 05 March 2024 19:54
To: Nikolaos Dandoulakis <nick.dan@xxxxxxxx>
Cc: ceph-users@xxxxxxx <ceph-users@xxxxxxx>
Subject: Re: Number of pgs
This email was sent to you by someone outside the University.
You should only click on links or attachments if you are certain that the email is genuine and the content is safe.
If you only have one pool of significant size, then your PG ratio is around 40 . IMHO too low.
If you're using HDDs I personally might set to 8192 ; if using NVMe SSDS arguably 16384 -- assuming that your OSD sizes are more or less close to each other.
`ceph osd df` will show toward the right how many PG replicas are on each OSD.
On Mar 5, 2024, at 14:50, Nikolaos Dandoulakis <nick.dan@xxxxxxxx> wrote:
Hi Anthony,
I should have said, it’s replicated (3)
Best,
Nick
Sent from my phone, apologies for any typos!
________________________________
From: Anthony D'Atri <aad@xxxxxxxxxxxxxx>
Sent: Tuesday, March 5, 2024 7:22:42 PM
To: Nikolaos Dandoulakis <nick.dan@xxxxxxxx>
Cc: ceph-users@xxxxxxx <ceph-users@xxxxxxx>
Subject: Re: Number of pgs
This email was sent to you by someone outside the University.
You should only click on links or attachments if you are certain that the email is genuine and the content is safe.
Replicated or EC?
On Mar 5, 2024, at 14:09, Nikolaos Dandoulakis <nick.dan@xxxxxxxx> wrote:
Hi all,
Pretty sure not the first time you see a thread like this.
Our cluster consists of 12 nodes/153 OSDs/1.2 PiB used, 708 TiB /1.9 PiB avail
The data pool is 2048 pgs big exactly the same number as when the cluster started. We have no issues with the cluster, everything runs as expected and very efficiently. We support about 1000 clients. The question is should we increase the number of pgs? If you think so, what is the sensible number to go to? 4096? More?
I will eagerly await for your response.
Best,
Nick
P.S. Yes, autoscaler is off :)
The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336. Is e buidheann carthannais a th' ann an Oilthigh Dhùn Èideann, clàraichte an Alba, àireamh clàraidh SC005336.
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
