[Bioclusters] notes and pictures from a "wet lab baby-biocluster" project

Chris Dagdigian dag at sonsorol.org
Wed Mar 23 10:06:59 EST 2005

Hi Matt,

I've used the KVM daisy chain stuff years ago on IBM rackmounts and 
enjoyed the convenience. It was not ordered for this product mostly 
because we had no plans to install even a single monitor/keyboard/mouse 
setup into the cluster at all. The goal was total headless management 
via remote power control and serial console redirection.

The choice of ATA storage was due to the fact that the cluster workflow 
was not going to be I/O bound at the fileserver level so high performing 
arrays were not needed. In my original plans I was looking at much 
higher end products from NetApp and BlueArc (initially a desire for 
extreme reliability was a factor) as well as other ATA arrays from 
companies like NexSan.

The specific choice of the Apple Xserve RAID was more tactical than 
anything else:

  - we've worked on lots of Xserve RAIDs w/ both Apple and Linux hosts

  - Apple's storage product is very price competitive even when compared 
to "whitebox" products yet comes with a good spare parts kit and service 

  - You can order it easily off a website, no fuss.

  - We have solid relationships with Apple techies, product managers and 
the sales folks. We knew if there was a product problem or shipping 
delay with the storage we could just "pick up the phone" to sort out any 
issues. *{very attractive when working under very tight delivery deadlines}

So no special reason why Apple storage was used other then the fact that 
we knew it would suit our needs, could be deployed rapidly and we had 
fallback resources on-call if something went awry.


Matt Hudson wrote:
> I'm interested by your choice of setup.
> We've got a cluster that bears a resemblance to this one, we have 30
> IBM e series IU duals (older X330 Pentium 3 machines), a Dell
> PowerVault 14x146GB U320 SCSI storage unit and a cheap Netgear switch.
> Two questions:
> 1) One thing I liked about the e series is the KVM daisy chain
> function; although I can see this is less useful for only 6 machines,
> I'm wondering if you chose not to buy it or if they've stopped
> offering this. It is great when it works, but it is very picky about
> keyboards, seems only to work with Microsoft branded ones (not IBM,
> strangely).
>  2) I take it that your application requires massive storage that
> isn't particularly fast, hence the array of ATA disks. Or is there
> something about the Apple that I'm missing? I would like to see how
> its performance over fiber channel compares with an Ultra 320 SCSI
> array controlled directly by the head node. 5.6 TB is an impressive
> amount of storage.

More information about the Bioclusters mailing list