CMOD Number of local caches

[wwwalton]

We run about 12 local caches, half about 300 GB and half about 800 GB on AIX 7 CMOD 8.5.  We have to grow it, are there any thoughts (or ROTs) about having small number of huge caches as opposed to a large number of smaller caches? These are not backed up but are replicated to a remote site.
Thanks,
-walt

[Justin Derrick]

I've always recommended that you have a higher number of caches in large systems, rather than a few massive filesystems.  My rationale is that if one cache filesystem goes offline (hardware issues, bad filesystem, SAN re-zoning, connectivity issues, etc.) that you lose a smaller percentage of the total, and that will likely mean that users will still get what they're looking for while the filesystem is being recovered in the background.

Of course, this presumes that your caches are substantially diverse enough to make a difference.  If 5 of your 10 cache filesystems are all on the same frame or RAID group, then a hardware failure on that system will take out 50% of your caches.

If you can't do that, then making sure your cache is on your most reliable/redundant disk (mirrored, RAID 5, RAID 5+0, etc.) is your best way forward.

Good luck!

-JD.

[Alessandro Perucchi]

I would just add from what Justin, rightly, just said, that if possible to have all cache FS at the same size is a best practice, since CMOD will check each directory in order to put the next file in round robin way.
If the FS are all of different size, then the round robin won't work as efficiently, and could use some FS more than others, defeating the purpose of the round robin, which ensure that evey FS is used as much as the others.

[Justin Derrick]

Excellent point, Alex...

If your caches are difference sizes, then some of them will represent a larger percentage of the total data cached -- and if that cache filesystem goes away, then you'll be missing a disproportionately larger percentage of your cache files.

The other thing I forgot to mention earlier was that in the scenario I mentioned earlier (10 caches, and 1 goes offline), if you have Tivoli Storage Manager ("TSM") installed, and are migrating new data to TSM at load time, end users probably won't even notice anything -- 10% of your retrieval requests will be served up by TSM if it can't find them in the cache.

-JD.