Originally posted by: martc
We are in the process of arranging an upgrade to the cache (not shared memory) to our 9980.
current - 20TB disk, 3G shared memory, 24G cache
proposed - 23TB disk, 3G shared memory, 26G cache
Our previous upgrades have been performed during outages or weekends when there was minimal disk access. As it is difficult getting outages and as upgrades are/maybe free (?) during prime shift, this time we are considering doing it online.
I am interested to hear of other's experiences of upgrading cache and if it is advisable doing it online.
There is a concern that perfomance will be affected. My understanding of the process (please correct me if I am wrong) is
1. flush (destage) cache in cluster 1
2. now only have half the memory for caching reads
writes in write-through mode
3. upgrade cache in cluster 1
4. bring cluster 1 cache into play
5. flush cache in cluster 2
6. again only half memory for caching reads and writes in write-through
7. upgrade cache in cluster 2
8. bring cluster 2 cache into play
The last time we did this it took 4-6 hours. Although there was minimal disk access.
Internally there is a concern that performance will be degraded during the flushing of the cache. And also the fact that there is a reluctance to pay extra for installation, the following is being suggested
1. day 1 (end of shift) - kick off flushing of cluster 1 cache and leave to run
2. day 2 (start of shift) - upgrade cache in cluster 1 and bring into play
day 2 (end of shift) - kick off flushing of cluster 2 cache and leave to run
3. day 3 (start of shift) - upgrade cache in cluster 2 and bring into play
Is it possible to run with different sizes of caches in each cluster? Suggestions and comments on the above would be appreciated as I do not feel comfortable with this suggestion. I would prefer to do it as a single job. After all, the 9980 is designed to allow upgrades,etc to be done online.
I am conscious that I am dragging this out a bit. Appologies. It is my understanding that when the cache in a cluster is upgraded, that cluster is not brought down and that the CHA ports/paths will still be active. Am I correct in thinking this?