# [sigh ….]

New unit being built for another customer [JackRabbit orders are multiplying like bunnies]

[root@jackrabbitm ~]# dd if=/big/big.file of=/dev/null  ...
10000+0 records in
10000+0 records out
83886080000 bytes (84 GB) copied, 51.8149 seconds, 1.6 GB/s
[root@jackrabbitm ~]# cat /proc/meminfo | grep MemTotal
MemTotal:     33011556 kB


Yes. We did just stream a file more than 2x the size of ram (32 GB) from disk. Yes, it sustained 1.62 GB/s.
Yes. We did this with 24 disks. Imagine what we could do with 48.

### 7 thoughts on “[sigh ….]”

1. 1.6 GB from 24 disks is about 68 MB per disk. that is to be expected and pretty normal.

2. This assumes that you have a RAID0. This unit is organized as 2 RAID6’s with 2 hot spares. This means 18 active drives.
Or 90 MB/s per drive.

3. sequential read or streaming with a decent off the shelf RAID card should be able to deliver 90+ MB/s esp. with the improved stride detection algorithms and nice fast cache in modern RAID cards.

4. We have tried a few, and haven’t seen that. Much closer to 50-60 MB/s in best efforts, 20-30 MB/s in worst efforts.
The caching on RAID cards is a double edged sword. Make it too small and it has no impact. Too large and you get a “sloshing” effect (which we do see in a number of large streaming workloads). Sadly, optimal cache sizing is somewhat of a black art … using some statistics …

5. Hmmm …. the phrase ‘stride detection’ rang a bell, I had heard it before … but didn’t know where. A fast google of ‘stride detection disk storage raid’ turns this phrase up in ZFS whitepapers and marketing documents.
Hmmm ….

6. ooh! Sun is scared of jackrabbit …

7. Not likely. No reason for them to be. Nothing to see here. Move along. Move along. 🙂
My point before is that people tend to adopt the jargon of where they work, or what they work with. This anonymous poster used language that I had heard in only one context, which I thought was odd.
I don’t mind (actually I encourage) anonymous postings as it enables people to be more honest in their assessments of things. There is a down side to that policy, but this wasn’t it.
They did trigger the memory, and it actually reminds me that we have some more testing to do. Specifically on OpenSolaris builds.