RAID controller without cache (performance)

[glideFX]

Hi,

I found a good deal for an LSI MegaRAID SAS 9240-8i which I plan to use for one RAID 10 volume.
Do you think the absence of the cache will be a performance issue?
How the cache impact on a streeping/mirroring array?

thanks

 

[BonzaiDuck]

Hi,

I found a good deal for an LSI MegaRAID SAS 9240-8i which I plan to use for one RAID 10 volume.
Do you think the absence of the cache will be a performance issue?
How the cache impact on a streeping/mirroring array?

thanks

Look at the cus reviews at the Egg and think about it. Will it work with your motherboard BIOS? I really wouldn't buy a RAID card of that capacity unless it were "Hardware RAID," and had onboard cache of -- what? -- 128MB? 256MB? If you're getting it for $50, maybe "take the chance" if you think you can use it. But this is a $300 controller card.

How many disks do you want, how many do you NEED? Just my thoughts.

 

[Soulkeeper]

You might consider linux software raid. With a reasonably fast cpu these setups tend to outperform most budget raid cards from what i've seen.

Opinions will differ on this, but from my perspective a hardware raid card is not worth the money unless it has battery backup and cache ... and you got tons of money.

Think of it this way, the hardware raid cards are usually a little cpu running at a few hundred MHz with some ddr1 or ddr2 memory. Sometimes it's just faster to rely on your host cpu, pcie, and disk controllers

 

[glideFX]

Hi, thank you for the answers.

I know what a (hardware) RAID controller is, I was just wondering if the performances of a cacheless card are so different from a cached controller if we speak about streeping and mirroring.
I know that for a RAID 5 / 6 the cache is more important.

Let's say I have a controller with 1GB DDR3 cache and a controller like the 9240 without any cache and I create a RAID0 volume on two WD RE4 2TB drives.
Where I should see improvements on the first controller? In the operations involving less than 1GB of data? In the random access? (small files)

If i copy a 4GB ISO from another drive (let's say an SSD) will the final speed be different on the two controllers? Will I have an initial boost on the cached controller?

Thank you

 

[Cerb]

Without cache, a RAID 10 should perform no better than the combination of the drives can allow. Basically, it should behave about like an onboard Intel RAID 10.

Parity RAIDs need cache because small writes may require reading before writing. Stripes in cache can be modified in the card's RAM, allowing just writes to be issued, if the LBAs being written to were recently read. The latency and IOPS can get bad. 0 and 1 type RAID arrays will never suffer from such problems, since there is no parity to calculate (they rely on the other drive holding good data, on an error).

 

[glideFX]

so the conclusion is, if it's a good deal and you are going to use it in a RAID 0/1/10 mode, buy it?

 

[Cerb]

Yes. RAID 10 with a cache could be faster, in that reads from the cache take no HDD IOPS away, but short of a very large cache (IE, SSD cachecade), or write cache (usually requires battery to enable), it won't make much difference, because the OS will do plenty of read caching on its own.

 

[gea]

Raid-Controller cache + BBU is mainly important to improve data security (avoid filesystem and raid-inconsistency on a crash during write) due to raid-hole problems.

Regarding performance a controller cache is quite useless.
Modern filesystem and operating systems (mainly ZFS but every new OS use multi Gigabyte of fast RAM as cache).

If you can use ZFS (BSD, Linux, Solaris) or btrfs or ReFS, such a controller without cache is perfect as a cache + BBU is not needed on a CopyOnWrite filesystem.

 

[Emulex]

with the 9240, esxi will throw latency warnings left and right, with a 9260/927x series with BBWC it will not throw latency warnings. No elevator write back caching will hurt latency big time!