The lower-cost controllers also offer RAID 10 but the S-series controllers must use system resources to manage RAID function, and the H200/SAS 6/iR have no cache.
What is the effect of no cache with the H200 controllers? I guess I don't know if a 64 meg cache even matters anymore with something like a 2 TB array, and Windows itself already does its own caching.
Having the controller try to randomly guess at what needs to be cached sounds mostly useless with such huge disk arrays. Now, if the cache size were scaled way up to match the array size so that you have say an 8-gig disk cache for a 2 TB array, that would make much more sense.
,
The main issue I know of, is that you probably cannot ever drop power on a no-cache RAID, even accidentally, because it means the disks go out of sync, and mirroring may need to be done to resync the mirror on reboot.
Or at least that's how it used to be with Novell Netware's software RAID. Every reboot after a bad shutdown or crash, the mirror is broken and it remirrors for an hour... meanwhile no redundancy...
That's also how it worked with a no-battery "LSI MEGARAID SATA150" controller I've used previously. If it don't shutdown clean, it declares the mirror bad and wants to remirror on restart.
You'd have to look up benchmark material for the benefits of RAID cache to see if it is something you would benefit from, but the PERC 6/i has a 256MB battery-backed cache (which I have heard of people upgrading to 512MB), and the H700 has up to 1GB cache, with the option of using an SSD for additional cache (CacheCade).
You won't find the kind of volatility with hardware RAID arrays that you do with software RAID (what you describe with the Netwares I've never seen anything like that on a Dell controller). I would imagine that whether there is cache or not, writes are nearly syncronous between the two RAID 1 drives, preserving the integrity between the two member disks.
I didn't look entirely through this article, but it looks like it might have some useful information.
The cache on the controller is not used to store frequently accessed data. The cache serves two purposes. It stores the current array configuration, and it acts as a buffer for data transmission.
Without cache the controller has to stop the flow of data until the drive has acknowledged that it is ready to receive. The controller will then receive the data from the OS and forward it on to the drive to be written. This process is very fast(microseconds) and can hardly be noticed.
Cache allows the controller to go ahead and receive data from the OS and place it into cache memory if the drive is not ready for it at that time.
Cache is the reason most RAID controllers are backed up by a battery. If there is a power loss and there is no battery backup to retain the cache memory then all of that data sitting in cache would be lost and cause file corruption. If the cache is backed up by a battery then the controller will complete writing that data to the drives when the drives come back online.
Also, just to clarify one thing ... the RAID cache does not hold RAID configuration settings. Those settings are stored in non-volatile memory on the controller (and the drives), so even if you unplug your server and disconnect the RAID battery to dump the cached data, you won't lose your RAID settings.
theflash1932
11 Legend
•
16255 Posts
2752
0
Posted February 24th, 2012 13:00
You'd have to look up benchmark material for the benefits of RAID cache to see if it is something you would benefit from, but the PERC 6/i has a 256MB battery-backed cache (which I have heard of people upgrading to 512MB), and the H700 has up to 1GB cache, with the option of using an SSD for additional cache (CacheCade).
You won't find the kind of volatility with hardware RAID arrays that you do with software RAID (what you describe with the Netwares I've never seen anything like that on a Dell controller). I would imagine that whether there is cache or not, writes are nearly syncronous between the two RAID 1 drives, preserving the integrity between the two member disks.
I didn't look entirely through this article, but it looks like it might have some useful information.