• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Next-Gen Velociraptors

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Originally posted by: taltamir
besides which, cache always runs out, cache helps with bursts followed by long inactivity, they don't help with anything sustained.
Click to expand...

How many 4KB random writes is your program going to issue in 10 seconds?

Entry level raid card with 128MB cache can buffer 32,768 4KB writes.

With a vertex retiring 1,536 4KB writes per second you'd need some silly stupid 4KB write sustained activity to saturate even a low level cached controller to cause stuttering.

I put 2GB in my areca controller because I could, not because I needed to. If I could put 8GB in it I would do it, but probably has no practical value to me as an end user at those levels.
 
but some controllers (cough jmicron cough) can't manage a single 4k write per second. sure you could keep rely on cache, but you get a larger and larger time period of unsafe data, with a 32,768 buffered writes it could have days of unwritten data in the cache that the computer thinks was written down; all lost when you shut off the pc... obviously that wouldn't happen in a well written cache algorithm as it stops caching when too many seconds/miliseconds have accumulated. resulting in the stutter despite having a massive cache.

So beyond a certain amount cache doesn't help anymore, only faster actual commit helps.
 
Faster commit comes with more spindles! Even though my STR is limited to around 900MB/S, adding more drives shortens flush times in full write back mode. This would be true with SSD's however we're going to see an new iop designed around SSD so they scale with at least 8 drives or more particularly for RAID5/6 scenarios. SSD's WILL fail just like a mechanical drive and when they do and your arrays aren't designed around high availability you will cry.
 
unless the ssd raid0 is just for running stuff really fast and they are all backed up nightly to an external spindle raid6 array 🙂

but yea, you are right.
 
Originally posted by: wired247
I've been doing some searching on google but I haven't heard (and couldn't find) anything about the next iteration of velociraptors. I assume they will be putting out higher density platters eventually, but when?
Click to expand...

I'm guessing the next Velociraptor will be a SSD.
 
Wow... That's big news that WD acquired SiliconSystems. I guess they really are entering the ssd market. Why wasn't that reported in dailytech or AT? That's big. Can't wait to see the result. I expect they'll be very professional about regression testing it like Intel (as opposed to ocz).
 
Originally posted by: Golgatha
Originally posted by: wired247
I've been doing some searching on google but I haven't heard (and couldn't find) anything about the next iteration of velociraptors. I assume they will be putting out higher density platters eventually, but when?
Click to expand...

I'm guessing the next Velociraptor will be a SSD.
Click to expand...

Silosaurus. :laugh:
 
Originally posted by: taltamir
but some controllers (cough jmicron cough) can't manage a single 4k write per second.
Click to expand...

Of course they can, it's just that once you exceed their internal buffer(16KB on the Jmicron), successive writes get very slow until the controller catches up.

 
@Golgotha: oops... I read dailytech on my blabkberry, and the page comes out all funny. I had looked at their homepage and didn't find that article until now. Thanks for the link.

Anyway, it's big news!
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top