Comments on: XIV Deep Dive Discussion http://blog.nigelpoulton.com/xiv-deep-dive-discussion/#utm_source=feed&utm_medium=feed&utm_campaign=feed with nigel poulton Tue, 24 Aug 2010 18:10:47 +0000 http://wordpress.org/?v=2.8.6 hourly 1 By: #EMCworld 2010 » XIV Recap http://blog.nigelpoulton.com/xiv-deep-dive-discussion/comment-page-1/#comment-755 #EMCworld 2010 » XIV Recap Wed, 05 May 2010 22:06:31 +0000 http://blog.nigelpoulton.com/?p=787#comment-755 [...] I'll send a PDF (provided there is some mechanism to decently print the Wave).  There was a podcast Nigel hosted last week that I participated in available on his podcast [...] [...] I'll send a PDF (provided there is some mechanism to decently print the Wave).  There was a podcast Nigel hosted last week that I participated in available on his podcast [...]

]]> By: IvanE http://blog.nigelpoulton.com/xiv-deep-dive-discussion/comment-page-1/#comment-748 IvanE Fri, 16 Apr 2010 06:22:49 +0000 http://blog.nigelpoulton.com/?p=787#comment-748 Not trying to be mean but couple of things I wanted to note on abviously most discussed part - double failures: - Nigel, you're not missing anything :) Highly utilized box under load will take over 30 minutes to rebuild after hard failure. IBM's own performance papers show exactly that. - There was a mention about contacting lab in case of double failure. Nice idea, but... XIV does not stop I/O in case of double failure. It's silent. So your applications will keep running and updating data until they actually hit lost area. In most cases this will lead to incosistency rendering your data useless. - There was a discussion about risk window. What always being overlooked is that you don't need two drives to fail - it's enough to have an independent failure of a module and one drive within 4-6 hours timeframe. You can't bring the module back until IBM's engineer comes around. Rebuild of module takes those 4-6 hours. An the box keeps running even though data is lost (see previous comment). - When talking about rebuild times in XIV as compared to traditional RAID, IBM never mentions that with RAID you risk one disk group, with XIV - the whole array. So, with XIV the probability might be less (due to faster rebuilds) but the impact in most cases is much, much higher. These are the reasons I voted "No, I don't like XIV, yes, I have tried it". Not trying to be mean but couple of things I wanted to note on abviously most discussed part – double failures:

- Nigel, you’re not missing anything :) Highly utilized box under load will take over 30 minutes to rebuild after hard failure. IBM’s own performance papers show exactly that.

- There was a mention about contacting lab in case of double failure. Nice idea, but… XIV does not stop I/O in case of double failure. It’s silent. So your applications will keep running and updating data until they actually hit lost area. In most cases this will lead to incosistency rendering your data useless.

- There was a discussion about risk window. What always being overlooked is that you don’t need two drives to fail – it’s enough to have an independent failure of a module and one drive within 4-6 hours timeframe. You can’t bring the module back until IBM’s engineer comes around. Rebuild of module takes those 4-6 hours. An the box keeps running even though data is lost (see previous comment).

- When talking about rebuild times in XIV as compared to traditional RAID, IBM never mentions that with RAID you risk one disk group, with XIV – the whole array. So, with XIV the probability might be less (due to faster rebuilds) but the impact in most cases is much, much higher.

These are the reasons I voted “No, I don’t like XIV, yes, I have tried it”.

]]> By: IvanE http://blog.nigelpoulton.com/xiv-deep-dive-discussion/comment-page-1/#comment-747 IvanE Fri, 16 Apr 2010 06:21:46 +0000 http://blog.nigelpoulton.com/?p=787#comment-747 Not trying to be mean but couple of things I wanted to note on abviously most discussed part - double failures: - Nigel, you're not missing anything :) Highly utilized box under load will take over 30 minutes to rebuild after hard failure. IBM's own performance papers show exactly that. - There was a mention about contacting lab in case of double failure. Nice idea, but... XIV does not stop I/O in case of double failure. It's silent. So your applications will keep running and updating data until they actually hit lost area. In most cases this will lead to incosistency rendering your data useless. - There was a discussion about risk window. What <strong>always</strong> being overlooked is that you don't need two drives to fail - it's enough to have an independent failure of a module and one drive within 4-6 hours timeframe. You can't bring the module back until IBM's engineer comes around. Rebuild of module takes those 4-6 hours. An the box keeps running even though data is lost (see previous comment). - When talking about rebuild times in XIV as compared to traditional RAID, IBM never mentions that with RAID you risk one disk group, with XIV - the whole array. So, with XIV the probability might be less (due to faster rebuilds) but the impact in most cases is much, much higher. These are the reasons I voted "No, I don't like XIV, yes, I have tried it". Not trying to be mean but couple of things I wanted to note on abviously most discussed part – double failures:
- Nigel, you're not missing anything :) Highly utilized box under load will take over 30 minutes to rebuild after hard failure. IBM's own performance papers show exactly that.
- There was a mention about contacting lab in case of double failure. Nice idea, but… XIV does not stop I/O in case of double failure. It's silent. So your applications will keep running and updating data until they actually hit lost area. In most cases this will lead to incosistency rendering your data useless.
- There was a discussion about risk window. What always being overlooked is that you don't need two drives to fail – it's enough to have an independent failure of a module and one drive within 4-6 hours timeframe. You can't bring the module back until IBM's engineer comes around. Rebuild of module takes those 4-6 hours. An the box keeps running even though data is lost (see previous comment).
- When talking about rebuild times in XIV as compared to traditional RAID, IBM never mentions that with RAID you risk one disk group, with XIV – the whole array. So, with XIV the probability might be less (due to faster rebuilds) but the impact in most cases is much, much higher.
These are the reasons I voted "No, I don't like XIV, yes, I have tried it".

]]> By: Hector Servadac http://blog.nigelpoulton.com/xiv-deep-dive-discussion/comment-page-1/#comment-730 Hector Servadac Tue, 23 Mar 2010 00:32:30 +0000 http://blog.nigelpoulton.com/?p=787#comment-730 Nigel... any chance you can transcript this? It's very difficult to hear Alan... Nigel… any chance you can transcript this? It's very difficult to hear Alan…

]]> By: Nigel Poulton http://blog.nigelpoulton.com/xiv-deep-dive-discussion/comment-page-1/#comment-729 Nigel Poulton Sat, 20 Mar 2010 15:56:16 +0000 http://blog.nigelpoulton.com/?p=787#comment-729 Hi SRJ, Thanks for your comment and Im gald we came across as balanced and lacking FUD. Id told the guys on the podcast that I expected it to run for ~30 minutes. Turned out to be about 50minutes so we didnt have time to mention everything. One other thing from the top of my head - Ive been told that you can grow volumes that are participating in live remote replication.... If you grow the source volume then remote volume is automatically grown! Sounds like a great feature to me. Id be interested to know what else you like about XIV. Nigel Hi SRJ,

Thanks for your comment and Im gald we came across as balanced and lacking FUD.

Id told the guys on the podcast that I expected it to run for ~30 minutes. Turned out to be about 50minutes so we didnt have time to mention everything. One other thing from the top of my head – Ive been told that you can grow volumes that are participating in live remote replication…. If you grow the source volume then remote volume is automatically grown! Sounds like a great feature to me.

Id be interested to know what else you like about XIV.

Nigel

]]> By: SRJ http://blog.nigelpoulton.com/xiv-deep-dive-discussion/comment-page-1/#comment-728 SRJ Sat, 20 Mar 2010 15:38:14 +0000 http://blog.nigelpoulton.com/?p=787#comment-728 Nice work guys...enjoyed the podcast.  I was impressed by the fairness and lack of FUD.  Techmute did a great job of summarizing our Google Wave on the subject... In response to the question - "What are the pro's of XIV?" - in addition to what was said, I would have just added a bit about the snapshot capability.  Their implementation is second-to-none in the industry and they do a very poor job of marketing this fact.  Like NetApp, there is no performance hit, but the XIV is not limited to 255 per volume.  Also, with XIV you can restore an older snapshot without destroying all of the subsequent snapshots, which is pretty cool. Nice work guys…enjoyed the podcast.  I was impressed by the fairness and lack of FUD.  Techmute did a great job of summarizing our Google Wave on the subject…
In response to the question – "What are the pro's of XIV?" – in addition to what was said, I would have just added a bit about the snapshot capability.  Their implementation is second-to-none in the industry and they do a very poor job of marketing this fact.  Like NetApp, there is no performance hit, but the XIV is not limited to 255 per volume.  Also, with XIV you can restore an older snapshot without destroying all of the subsequent snapshots, which is pretty cool.

]]>