<html><head><meta http-equiv="Content-Type" content="text/html; charset=iso-8859-15" /><meta name="Generator" content="Microsoft Exchange Server" /><style><!-- .EmailQuote { margin-left: 1pt; padding-left: 4pt; border-left: #800000 2px solid; } --></style></head><body>Hi Peter,<br>
<br>
I've always been under the impression that this is a ball park figure that changes some of the on disk data structures to help parallel access to the filesystem I try to estimate the number of clients and then add ~10% (depending on the cluster size ofc). <br>
<br>
I have tested the default 32 vs 200 on a previous cluster however didn't find a difference in performance when testing up to 50 clients concurrently with IOR random and sequential. Maybe the difference is more subtle that just throughput?<br>
<br>
What I've always found strange is that the value can be changed on the filesystem however I don't believe this change effects an already created filesystem.<br>
<br>
-- Lauz<br><br><div class="gmail_quote">On 21 October 2016 21:35:15 BST, Peter Childs <p.childs@qmul.ac.uk> wrote:<blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;">
<!-- converted from text -->
<div><br />
Reading through them and we'll worth read.<br />
<br />
How important is setting the correct cluster size at file system creation time? Ie with "mmchfs -n 256" ie how much band of error can you get away with?<br />
<br />
We have a cluster of 240 nodes that was setup with the default 32 setting, it's now about to grow to ~300 nodes. Is this likly to causing as an issue, which is difficult to fix. The manual says it can be adjusted but this white paper suggests not.<br />
<br />
Fortunately were also migrating our storage to new hardware, so have a good opportunity to get the setting right, this time around.
<br />
<br />
Has anyone got any stats on the benefits of getting it "right" vs getting it wrong.<br />
<br />
<br />
<br />
<br />
Peter Childs<br />
Research Storage<br />
ITS Research and Teaching Support<br />
Queen Mary, University of London<br />
<br />
<br />
---- Andreas Landhäußer wrote ----<br />
<br />
</div>
<font size="2"><span style="font-size:10pt;">
<div class="PlainText">On Fri, 21 Oct 2016, Laurence Horrocks-Barlow <laurence@qsplace.co.uk> wrote:<br />
<br />
Found it!<br />
<br />
thanks Yuri, for the two very interesting papers about the Metadata and <br />
replication.<br />
<br />
We always used the rule of thumb 5% for metadata, since we are having <br />
separate devices for metadata, and tiny fast disks aren't available <br />
anymore, we are getting a bunch of larger fast disks. We never experienced <br />
a problem with the (more or less) amount of metadata ...<br />
<br />
Andreas<br />
<br />
<br />
> Right down the bottom of the page under attachments.<br />
><br />
> -- Lauz<br />
><br />
> On 21 October 2016 07:43:40 BST, "Andreas Landhäußer" <alandhae@gmx.de> wrote:<br />
>><br />
>> Hi Yuri,<br />
>><br />
>> Arrg, can't find them, page has been last updated on Aug, 18 by<br />
>> JohnTOlson, maybe its internal and not open for the public?<br />
>><br />
>> Ciao<br />
>><br />
>> Andreas<br />
>><br />
>> On Fri, 21 Oct 2016, Yuri L Volobuev <volobuev@us.ibm.com> wrote:<br />
>><br />
>>><br />
>>><br />
>>> Esteemed GPFS and Spectrum Scale users,<br />
>>><br />
>>> For your reading enjoyment, two new whitepapers have been posted to<br />
>> the<br />
>>> Spectrum Scale Wiki:<br />
>>><br />
>>> Spectrum Scale: Replication in GPFS<br />
>>> Spectrum Scale: GPFS Metadata<br />
>>><br />
>>> The URL for the parent page is<br />
>>><br />
>> <a href="https://www.ibm.com/developerworks/community/wikis/home?lang=en#!/wiki/General%20Parallel%20File%20System%20">
https://www.ibm.com/developerworks/community/wikis/home?lang=en#!/wiki/General%20Parallel%20File%20System%20</a><br />
>>> (GPFS)/page/White%20Papers%20%26%20Media<br />
>>><br />
>>> The two .pdf documents are accessible through the Attachment section<br />
>> at the<br />
>>> bottom of the page. Unfortunately, dW "spam prevention engine" does<br />
>> a very<br />
>>> good job preventing me from "spamming" the page to actually add<br />
>> links.<br />
>>><br />
>>> Best regards,<br />
>>><br />
>>> Yuri<br />
>>><br />
>><br />
>> --<br />
>> Andreas Landhäußer +49 151 12133027 (mobile)<br />
>> alandhae@gmx.de<br />
>><br />
>> ------------------------------------------------------------------------<br />
>><br />
>> _______________________________________________<br />
>> gpfsug-discuss mailing list<br />
>> gpfsug-discuss at spectrumscale.org<br />
>> <a href="http://gpfsug.org/mailman/listinfo/gpfsug-discuss">http://gpfsug.org/mailman/listinfo/gpfsug-discuss</a><br />
><br />
><br />
<br /></div></span></font></blockquote></div><br>
-- <br>
Sent from my Android device with K-9 Mail. Please excuse my brevity.</body></html>