Comments on: A Big SSD Vendor Begs To Differ http://storagemojo.com/2006/10/23/a-big-ssd-vendor-begs-to-differ/ Data storage info & analysis Wed, 09 Jul 2008 12:34:01 +0000 http://wordpress.org/?v=2.5.1 By: KirkH http://storagemojo.com/2006/10/23/a-big-ssd-vendor-begs-to-differ/#comment-129741 KirkH Fri, 05 Oct 2007 18:18:31 +0000 http://storagemojo.com/?p=284#comment-129741 Flash has the advantage in size but what if my whole DB fits in a few GB? I can get a small DDR based system (hyperdrive) for a fraction of the cost of the TMS beasts. Also, considering the typical replication setup wouldn't flash's poor write performance cease to be an issue? Just take it out of the list of available servers while you do the update. I'm just guessing but wouldn't the random read IOPS more than make up for the fact that write sucks? Maybe a cheap RAM based Hyperdrive4 as the master that handles writes and a bunch of flash drives on the slaves. Flash has the advantage in size but what if my whole DB fits in a few GB? I can get a small DDR based system (hyperdrive) for a fraction of the cost of the TMS beasts.

Also, considering the typical replication setup wouldn’t flash’s poor write performance cease to be an issue? Just take it out of the list of available servers while you do the update. I’m just guessing but wouldn’t the random read IOPS more than make up for the fact that write sucks?

Maybe a cheap RAM based Hyperdrive4 as the master that handles writes and a bunch of flash drives on the slaves.

]]> By: YC http://storagemojo.com/2006/10/23/a-big-ssd-vendor-begs-to-differ/#comment-95886 YC Sat, 14 Jul 2007 19:53:03 +0000 http://storagemojo.com/?p=284#comment-95886 classical example of disruption at work - and there is no question who will eat who's lunch later ;) classical example of disruption at work - and there is no question who will eat who’s lunch later ;)

]]> By: Kevin Closson http://storagemojo.com/2006/10/23/a-big-ssd-vendor-begs-to-differ/#comment-10001 Kevin Closson Mon, 20 Nov 2006 22:35:03 +0000 http://storagemojo.com/?p=284#comment-10001 Is a TMS really an SPOF? Sure, disaster, but they are very redundant along the lines of power and so on. I am testing my old Imperial Megaram 5000 (8port 1Gb FC Solid State) side by side against the modern TMS and I'll say that it is much better than just the simple 1Gb->2Gb expectation... Also, TMS resells PolyServe in which case is it quite simple to make a clustered NAS in front of TMS to provision direct I/O NFS access over, say, triple bonded GbE to serve as an "accelerator" if you will for transaction logs or temp sorting space...all without the connectivity and provisioning difficulty associated with FCP. It is an interesting approach to getting some SSD to a lot of servers--sans LUN masking and switch zoning... I think I've blogged that one at http://kevinclosson.wordpress.com/ Is a TMS really an SPOF? Sure, disaster, but they are very redundant along the lines of power and so on. I am testing my old Imperial Megaram 5000 (8port 1Gb FC Solid State) side by side against the modern TMS and I’ll say that it is much better than just the simple 1Gb->2Gb expectation…

Also, TMS resells PolyServe in which case is it quite simple to make a clustered NAS in front of TMS to provision direct I/O NFS access over, say, triple bonded GbE to serve as an “accelerator” if you will for transaction logs or temp sorting space…all without the connectivity and provisioning difficulty associated with FCP. It is an interesting approach to getting some SSD to a lot of servers–sans LUN masking and switch zoning…

I think I’ve blogged that one at http://kevinclosson.wordpress.com/

]]> By: Miro http://storagemojo.com/2006/10/23/a-big-ssd-vendor-begs-to-differ/#comment-7038 Miro Tue, 24 Oct 2006 05:05:38 +0000 http://storagemojo.com/?p=284#comment-7038 Not every enterprise is ready to trade redundancy, failover and price point for speed! While there are some places where high speed is needed, in most cases a distributed cluster solution is a much better choice. Google is a proof - how with cheap computers and distributed FS one can build the fastest search in the world - without using fancy storage. Everything fails sooner or later - nothing is 'rock solid' says Google... Now, if Sun can make the ZFS cluster capable/distributed... :) Every time when a sales manager is explaining to me how 'rock solid' solution centralized storage is, I can't stop thinking about Titanic... Not every enterprise is ready to trade redundancy, failover and price point for speed!

While there are some places where high speed is needed, in most cases a distributed cluster solution is a much better choice.

Google is a proof - how with cheap computers and distributed FS one can build the fastest search in the world - without using fancy storage.

Everything fails sooner or later - nothing is ‘rock solid’ says Google…

Now, if Sun can make the ZFS cluster capable/distributed… :)

Every time when a sales manager is explaining to me how ‘rock solid’ solution centralized storage is, I can’t stop thinking about Titanic…

]]> By: Richard http://storagemojo.com/2006/10/23/a-big-ssd-vendor-begs-to-differ/#comment-7015 Richard Tue, 24 Oct 2006 03:01:32 +0000 http://storagemojo.com/?p=284#comment-7015 Robin, Texas Memory SSD is perfect for high speed recording of internet 'backbone' traffic (security) or similar applications where speed is the #1 requirement and the cost becomes a secondary issue. In a typical HA solution the cost will double, as this product is a 'single point of failure' and replication is required. Robin,
Texas Memory SSD is perfect for high speed recording of internet ‘backbone’ traffic (security) or similar applications where speed is the #1 requirement and the cost becomes a secondary issue.

In a typical HA solution the cost will double, as this product is a ’single point of failure’ and replication is required.

]]>