cache size on XP128

[cart]

Hi,

We want to buy an XP128 with 1 TB of disks (9 AG with raid 5). How much cache should I get? I was looking for 4 GB, HP says 6 GB is a minimum which doesn't make sence to me....

Thanks for your help,

Sundance

[Peter Mattei]

Hi

You are obviously talking about 9 Array Groups of 36GB 15kRPM drives. This will give you app. 950GB usable space in Raid5.
So we are talking about very high performance disks and backend.
Since the XP128 is a cache centric machine cache sizing is vital to performance.
For your configuration 6GB is the absolute minimum. For very high write IO rates we recommend 10GB.

Enjoy your XP128. It is the fastest and most advanced storage array currently available on the market.

Peter

[Vincent Fleming]

Hi Sundance;

For XP's, the recommendation is to front load the cache - buy a lot to start, then a little as you expand.

The reason for this is the low number of drives will throttle your performance, and the large cache makes up for this. With more drives (5TB, for example), you don't need as much cache per GB because you have so many more spindles to balance your load on.

Here's the minimum reqs for the XP128:

up to 160GB = 2GB cache
160 to 700GB = 4GB
700 to 1350 = 6GB
1350 to 2752 = 8GB
2752 to 5517 = 10GB
5517+ = 12GB

So, the first 2GB of cache is for 160GB, the second handles 540GB, the third, 650GB, the fourth 1400GB, the fifth, 2765GB. More spindles means you need less cache per GB of data.

I hope this helps;

Good luck!

[Dave Wherry]

I would certainly go with the 6GB. I once worked in a shop where the "Senior" System Architect lead the evaluation of an EMC frame. He claimed it did not perform any better than the JBOD drives that were on the host. He rejected it and it was removed. I was not involed so I assumed the EMC was overhyped.
Later that year I moved to a shop where we had two EMC frames. Once I understood how these big Integrated Data Cache (IDC) frames worked, I realized why he rejected it. Rather than let the frame do its' job, use that cache, he had configured it with more drives to spread across more spindles, the old tried and true method, and very little cache. He essentially neutured the frame because he wanted to stick with the old methodology. I'm surprised EMC let him do that. Just as HP is telling you to go with 6GB of cache, EMC has their rules for configuring a frame. They goofed up.
I've told several people there is a bit of a disconnect you need to do from the old, tried and true, method of spreading your load across many spindles. With an IDC frame, be it XP, EMC, HDS..., all writes go to cache and are acknowledged to the host so it may proceed with the next operation. Later on they will be written to disk. Many reads are serviced from cache as the data may already be in cache, or the frame may perform a read ahead, anticipating what data the host will need next. While more spindles should not hurt performance, big cache is where you truly gain performance.
If you are looking at an XP just for the storage capacity you might want to look at a SAN solution and save some money. If you are looking for capacity and excellent performance, give it cache. Of course it will cost cash.