From the SAN Guide:
Caching
Though ESX Server systems benefit from write cache, the cache could be saturated
with sufficiently intense I/O. Saturation reduces the cache's effectiveness.
Because the cache is often allocated from a global pool, it should be allocated only if
it will be effective.
A read-ahead cache may be effective for sequential I/O, such as during certain
types of backup activities, and for template repositories.
A read cache is often ineffective when applied to a VMFS-based volume because
multiple virtual machines are accessed concurrently. Because data access is
random, the read cache hit rate is often too low to justify allocating a read cache.
A read cache is often unnecessary when the application and operating system
cache data are within the virtual machine's memory. In that case, the read cache
caches data objects that the application or operating system already cache
From reading the above, it would appear that Read Cache for VMFS Luns is almost pointless, have many people disabled Read Cache (or minimised) on VMFS Luns? and what has your expericne been?
I'm running an EVA8000.
Cheers
Altonius
Try it out... - it's only a few mouse clicks in Command View EVA.
(We have a customer whose EVA had a problem with read caching / prefetching which cause massive performance problems.)
Things are ticking along OK at the moment, as we only have minimal load on the EVA. I was wondering what sort of load you were experiencing before you noticed the performance problems.
Altonius
I don't have any numbers, sorry. Every system is an individual and behaves differently.
The system in question provides storage for a number of servers running ESX, Tru64 Unix and Windows. The customer does provide services/hosting and keeps an eye on the performance of his systems, so he immediately noticed the problem.