VMware Horizon Community
Rkelly
Contributor
Contributor

Anyone try Atlantis Computing's iLiO with VMware View yet?

http://www.atlantiscomputing.com/

This is some pretty interesting technology. Of fcourse the catch is "does it work?"

Apperently unidesk.com has the same technology but now working product.

Reply
0 Kudos
17 Replies
Rkelly
Contributor
Contributor

Here is a great overview of what these technologied do:

http://www.brianmadden.com/blogs/brianmadden/archive/2009/02/18/brian-dump-atlantis-computing-hopes-...

Also, correction on my original post. I meant to say that Unidesk does not have a working product yet. We will have to wait and see what they deliver.

Reply
0 Kudos
patten_aaron
Enthusiast
Enthusiast

I am exploring using ILIO on an 8000 seat View 3 deployment. As part of that discussion, I am meeting with them today to discuss the ILIO product in detail and will post thoughts on the conversation after the meeting.

~Aaron Patten

~Aaron Patten Please award points for helpful answers. I got kids to feed 🙂
Reply
0 Kudos
Rkelly
Contributor
Contributor

Hi Aaron,

Yes, please post your findings. This looks like very ground breaking technology that would make for a very nice Vmware View deployment. The gold disk guys at my company would love this if it works.

Reply
0 Kudos
patten_aaron
Enthusiast
Enthusiast

Some details about ILIO from my discussion with them:

ILIO actually has a family of products including Aggregator and Controller. These two work together to accelerate and cache IO "between" ESX and attached storage. The controllers are VMs that are given around 3 to 4 GB of RAM and they use that as a sort of RAM disk for IO local to that ESX server. The Aggregator controls multiple Controllers and is usually asigned 16 GB of RAM or so. Supposedly this allows them to mitigate read heavy IOPS storms like boot/logon/logoff/virus scan type of activity.

In addition, there is a lifecycle management portion of the product which allows the type of change controll and image management shown in the Brian Madden video that was linked earlier. This can be deployed separately from the Aggregator/Controller products. Refresh and recompose operations no longer need to be run on a regular basis to patch images as this can be done on the fly as demonstrated in the videos. Also, there is a method that allows the shrinkage of the delta disk usage, so it's possible to reclaim space as users delete data from their desktops. (<- if this works as advertised, it is a HUGE plus when working with arrays that support Thin provisioning!).

That's all I have for now. Have a demonstration with them on Monday.

~Aaron Patten

~Aaron Patten Please award points for helpful answers. I got kids to feed 🙂
Reply
0 Kudos
jjgunn
Enthusiast
Enthusiast

"Have a demonstration with them on Monday."

Wondering how the demonstration went? Let us know if you can. Thank you

Reply
0 Kudos
patten_aaron
Enthusiast
Enthusiast

In the process of deploying the product now and will do some functionality and performance testing with the product at that time. I won't be able to give exact numbers, but should have a general feel for performance/features by the end of the week.

~Aaron Patten

~Aaron Patten Please award points for helpful answers. I got kids to feed 🙂
Reply
0 Kudos
jjgunn
Enthusiast
Enthusiast

Aaron, how was the install? Easy to follow and perform? difficult?

Let us know your findings. Thank you

Reply
0 Kudos
patten_aaron
Enthusiast
Enthusiast

So far, the layering technologies are very nice in the little I've played with them. The ability to make changes to the images (more or less) on the fly is very powerful. The install is not too involved either but the documentation could use a little clarification in some areas. Basically you export and NFS mount on your NAS device (using Celerra in my environment) and you mount this to the ILIO devices (virtual appliances). The ILIO devices in turn export this same share which you mount to the ESX hosts so that the ILIO sits in-line with your storage so that it can cache IO requests from the hypervisor which should increase performance and reduce IO load on the storage array.

I'm having some issues with the performance so far but we should get that licked soon.

~Aaron Patten

Please award points for helpful answers. I got kids to feed Smiley Happy

~Aaron Patten Please award points for helpful answers. I got kids to feed 🙂
nreineke
Contributor
Contributor

I just came across this post, I am the Product Manager at Unidesk, and would be glad give you a live demo of our working product if you are interested. Additionally, we are ramping up our beta program, and always welcome additional participants.

If this is of interest, please send me a note: Nicole_Reineke @ unidesk.com

Reply
0 Kudos
OWScott
Contributor
Contributor

Aaron, any update on this from your usage over the last couple months? This looks like a compelling product so I'm curious to know if you got the performance issue resolved easily and how it's done for you since.

Thanks!

Scott

Reply
0 Kudos
patten_aaron
Enthusiast
Enthusiast

Scott,

It is indeed a promising product family. Unfortunately my testing took a back seat to some work I did for VMworld and other priorities are keeping it there for the time being. If I had to pick one component of the product to call out I'd highlight the lifecycle management portion. The true inheritance model they provide makes desktop management much easier that it was previously.

~Aaron Patten

Please award points for helpful answers. I got kids to feed Smiley Happy

~Aaron Patten Please award points for helpful answers. I got kids to feed 🙂
Reply
0 Kudos
OWScott
Contributor
Contributor

Hi Aaron,

Thanks for the reply. It does seem to have a lot of potential. I'm trying to get in touch with their sales or tech team to discuss details and hopefully demo it myself.

Scott

Reply
0 Kudos
arjunvenkatrama
Contributor
Contributor

Hi Scott

My name is Arjun Venkatraman and I'm the lead Solutions Architect for Atlantis Computing. I would be happy to give you a demo of our products and work out how to get it into your hands for an evaluation. You can reach me at arjun@atlantiscomputing.co.

Look forward to hearing from you

Regards

Arjun Venkatraman

Reply
0 Kudos
hmartin
Enthusiast
Enthusiast

I know this post is quite old, but wondering what, if anything, you did with ILIO.

Reply
0 Kudos
jeffkeyes
Contributor
Contributor

I just posted a demo of it here:  Atlantis USX Demo - VDI Performance Before and After on Vimeo

(Atlantis USX is the next generation of Atlantis ILIO with broadened use cases and capabilities beyond just VDI.  Massive boost for database, Exchange and other applications)

Jeff

jeffk@atlantiscomputing.com

Reply
0 Kudos
mbarnesotn
Contributor
Contributor

Good afternoon,

We are currently using Atlantis ILIO with View 6.  It does work as advertised.  Very speedy response from desktops (don't have numbers right off).  However, I would like to mention a few items that have now made us wonder if it is worth it.  BTW, we have 3 hosts for 100 desktops and we are using non-persistent linked clones.

1.  This one is in my mind the biggest flaw and reason we are reconsidering.  When we have to take a host down or if something happens and a host goes down, we have about an extra hour of images replicating before the desktops on that host are available to be used.  This is due to the replication of the images from disk to memory.  If you calculate this extra time in your maintenance, it is not a big deal.

2.  I love command line administration to systems.  Powershell and bash are my friends like Jack and Jim.  However, the administration of the ILIO appliances is just a bit more than I can remember.  I'm sure it is my understanding of the system and what it is doing.  I believe this would not be a problem if they had a one sheet that explained the basics.  I asked and was told they did not have one.

     a.   When you recompose a pool you should go into the appliance (one per host) and make sure no older images are kept.  This is just to help maintain the drive space and performance that the desktops use. We have had images there we had to manually remove.

     b.   This is not a problem but wanted to point this out.  Atlantis ILIO support staff have been very help and are top notch.  I usually get a 15 min cal back response and the techs have always been very helpful and willing to work with you.

3.  HA is not your friend so keep an eye on the configuration.  We had a vendor that turned on HA during a checkup.  We did not realize this at the time and when a host went down the ILIO Appliance moved.  This caused extra time recovery.  This is not necessarily a problem with ILIO but it needed to be pointed out.

4.  Resource (memory) hog.  We have 3 hosts with 256 GB of ram. In theory we should be able to have about 50 desktops per host.  With ILIO, we are limited to about 33.  We knew there would be a decent chuck of overhead but this surprised us.

I hope this helps.

Mike

Reply
0 Kudos
jeffkeyes
Contributor
Contributor

Just a few points of clarification (as things have improved with the introduction of Atlantis USX).  I'll add feedback from our team inline 

>>1.  This one is in my mind the biggest flaw and reason we are reconsidering.  When we have to take a host down or if something happens and a host goes down, we have about an extra hour of images replicating before the desktops on that host are available to be used.  This is due to the replication of the images from disk to memory.  If you calculate this extra time in your maintenance, it is not a big deal.


Obviously this is the diskless version of ILIO. On reboots yes – there will be add-on time for the contents of the disk (VM template images) to be copied back into its primary storage here in this model which is  RAM. The speed of the underlying disk will be a function of how fast the system can recover. That being said, there is an option to  switch to a Simple Hybrid Model (or traditionally referred to as Disk-backed ILIO) with low cost flash or spinning disk-fronted by memory as a performance tier. Here the recovery time window will reduce to seconds instead of minutes. That should be able to get you past this concern whilst keeping a high performance during operations as ILIO can take advantage of memory or flash to provide you the performance boost that you are currently experiencing in the pure in-Memory model.



>>2.  I love command line administration to systems.  Powershell and bash are my friends like Jack and Jim.  However, the administration of the ILIO appliances is just a bit more than I can remember.  I'm sure it is my understanding of the system and what it is doing.  I believe this would not be a problem if they had a one sheet that explained the basics.  I asked and was told they did not have one.

     a.   When you recompose a pool you should go into the appliance (one per host) and make sure no older images are kept.  This is just to help maintain the drive space and performance that the desktops use. We have had images there we had to manually remove.

     b.   This is not a problem but wanted to point this out.  Atlantis ILIO support staff have been very help and are top notch.  I usually get a 15 min cal back response and the techs have always been very helpful and willing to work with you.


Dating back to the manner in which ILIO was architected, it allowed a lot of latitude for system admins, script gurus to automate not just operations but automate installation and configuration elements of this technology in many different ways. One of the examples JPMC have been able to achieve a ridiculously high degree of automation with Diskless ILIOs (ILIO OVF to Working state to Automated Fast clone desktop composition to Reboot Workflows). Powershell,bash,Perl are all possible platforms that can be leveraged to automate required automations, health checks, recovery and repair flows so that it can be effectively weaved into a customized environmental needs. We will need to get you to speak to a resource from our Sa community who has good handle on View – PowerShell and work through your needs.



>>3.  HA is not your friend so keep an eye on the configuration.  We had a vendor that turned on HA during a checkup.  We did not realize this at the time and when a host went down the ILIO Appliance moved.  This caused extra time recovery.  This is not necessarily a problem with ILIO but it needed to be pointed out.


With regards to HA, its important to point out some best practices and techniques that can make the automation of migration more straight forward than what you have experienced. I would suggest going through the blog posted here http://community.atlantiscomputing.com/blog/Atlantis/February-2013/Desktop-Virtualization-Best-Pract... . You want to check your HA settings against notes made here, perhaps there are some more enhancements that can help you improve the overall experience of failure. Another option in a 3 or 4 Host environment is to leverage the Pooled in-Memory model or HyperConverged model with USX ( Newer more advanced technology that gives you the option of pooling different tiers of local resources between Host servers) which has an inherent HA and resilient model independent of the Hypervisor Resiliency services. The overall recovery time here should be consistently in a few seconds even more so when you are running non-persistent target workloads.



>> 4.  Resource (memory) hog.  We have 3 hosts with 256 GB of ram. In theory we should be able to have about 50 desktops per host.  With ILIO, we are limited to about 33.  We knew there would be a decent chuck of overhead but this surprised us.


It will interesting to figure out the size of the ILIO appliance is deployed here and what percentage of storage savings you are experiencing (we should be seeing fairly large deduplication % here). Yes, ILIO and USX does take up memory resources – depending on the model you are using there is varying levels of resources taken up with in-Memory which I believe is what you are running taking up maximum resources as compared to the other alternative USX, ILIO models that take a maximum of 5% of the memory resources.  There are mechanisms and ways even in the current model of your deployment to maximize and potentially get close to 50 desktops per host with the best performance.A more closer look at your environment can allow us to make some tweaks to the size of the data store deployed on a per host basis.



jeffk@atlantiscomputing.com


Reply
0 Kudos