Double-Take’s Flex product is like Citrix Provisioning Server without Citrix

Ironically, the line " and now they’re sort of a solution looking for a market"  was also in my (unreleased) white paper about Ardence that I was writing before they were purchased by Citrix.  

I was working with them and had the paper mostly complete when they asked me to retest with their new version about to come out, and then they suddenly went quiet on me.  Two months later Citrix made the announcement.

But yeah, assuming no patent like issues are in the way, there is no reason that many third party companies couldn't build this kind of solution.  Now that you brought this up, I am surprised that some of the other usual third party folks in this space don't also have something like this.

Brian - you may have viewed some older pricing information. Double-Take no longer charges for the Management Server. That $995 price is now $0.

You get the Flex Management Server and Flex Storage Server at no charge with the purchase of any Flex Cliet (Desktop Agent, $95 or Server Agent, $395.

Interesting.  We are considering an upgrade of all of our Enterprise licenses to Platinum to include the PVS licensing; however the DT option seems far more cost efficient.  Yes, there are other benefits to Platinum, but PVS was the main reason, and we don't use the CAG or EasyCall.  Edgesight Advanced can be added for very little.

So, for customers that want PVS but will not utilize all the features of a Platinum license, is the DT approach a more cost efficient option?  DT doesn't charge for support calls, as that is part of the maintenance agreement, and is less expensive per server.  What is the upside to Citrix PVS if DT Flex can do the same thing at a fraction of the cost?

Is anyone using Flex or done a comparison to PVS?  What does Citrix have to say about the competition?

My question is about ISCSI on the Desktop networks.  Maybe I am wrong, but what kind of impact does that have.  We plan for iSCSI in the server networks, but do people plan for it in the desktop networks?  Does it really matter?

Quote

"The only real downside to Citrix owning the product is that they’ve rolled it into their various XenDesktop, XenApp, and XenServer Suites—you can’t buy it as a standalone product anymore:"

Actually, you can. Citrix still offers Provisioning Services as a standalone product.  They just don't make a big deal about this.  You can order "Provisioning Services for Desktops" or "Provisioning Services for Data Centers" depending on what you want to use it for.

You're right, they don't make it easy.  

For Datacenters:

www.citrix.com/.../product.asp

Can't find a separate product page for the Desktop side.  Looks like it's just bundled in with XenDekstop, but hard to say.

For Joe Dropkin: you don't have to go to Platinum just to get PVS.  You can buy it on its own.  My current client just did this, and is running it in production now.

"Code quality and support"....LOL oh that's rich.

Re: Bryan Mann, Q on iSCSI on desktop networks:

Regarding the boot process:

·         2008 takes approximately 120 MB to boot (or to be streamed down from server) , XP, about 75 MB to boot from point of contact to iSCSI target to a Windows login prompt.

The largest any test of a simultaneous boot was done by a customer of ours in Japan. The test consisted of

·         205 Client PCs (PC running XP SP2, 700 MB RAM, 100 mbps NIC, 8 GB boot disks.)

·         iSCSI target was our own, running on 2003 R2, Xeon 2.33 GhZ, 3 GB RAM, Neterion Xframe 10 GbE NICs x 2, disks were SSD

·         our Flex Server, was running on 2003 R2, Celeron 2.93 GhZ, 512 MB RAM, Intel PRO/1000 NIC

Avg. time to boot was 33. 25 seconds. All clients booted successfully.

Post-boot:

NetApp and Intel state that the only concern for bandwidth utilization in boot-from-SAN implementations is during the boot process itself - never post boot.

b)  Intel Information Technology, Streaming and Virtual Hosted Desktop Study, 2008:

·         "Streaming promises to be a very efficient model, with very low server disk and processor utilization. Over time, streaming performs efficiently on the network, comparable to locally installed OS and applications."

NetApp white paper on "SAN Booting with Windows Server Operating Systems" from January 2008 states that "once the boot process is completed and applications started, steady state iSCSI activiy to the boot drive was less than 2 MB per minute."

We have some clients that have been booting  desktops in educational, financial, defense and corporate areas via iSCSI  - some for close to 4 years now, clients numbering in the 1000's - and these are on 10/100 networks. iSCSI itself for the desktop has not seemed to have been an issue.

@Steve, thanks.  I know it is not much once up and running, but "boot storms" are more likely in a desktop scenario than server.  I suppose you can implement an "infrastructure" per VLAN to avoid the storm of a bunch of people coming in and turning on @ 8 am.  I guess the same issues apply to the "UDP network" with PVS.

@Brian - There isn't a public link that I know of that shows the SKUs for Provisioning Server.  However, it's on the price list that Citrix provides to partners.

I believe list price for Citrix Provisioning Server for Desktops is $150.00 per connection.

@ all - Just thought I would 'pop in' on this excellent item / discussion. We've been users of DT Flex / Emboot Netboot/i for over three years now, choosing to PXE boot servers off iSCSI to Lefthand (now HP Lefthand) SANs. When we first told people what we were doing, they thought we were pretty crazy. "You are going to boot your servers off that WIRE?". It's still a concern people have today. They feel an almost gravitational force which makes them want to stick 2 X RAID 1 boot, and 3 X RAID5 disks in all their servers.

But our intention was to keep the bare metal separate from the OS and we stuck to that dream.

So we're more focused on Flex for servers than desktop  (at the moment) but the point of this contribution is to make the point that - as a long term user -that this stuff is SUPER reliable.

I can remember when I set up my first netboot setup (2 X SAN, 5 X diskless nodes, switches, management server etc) I was so confident in the technology, I drove 80 miles to the datacenter without a keyboard, screen or mouse (the DC I was in didn't have crash tables). I wired it all up. Switched on the SANs, switched on the nodes, in 10 minutes I was done - walked back to my car and went back to the office to enjoy my work. There was never a thought that it wouldn't be greeted by anything else but a Windows 2003 server login screen!

So just in case anyone was concerned about reliability... don't be!

Jeff

Fortunately, Citrix PVS and Flex are not the only OS/Disk streaming products on the market.

There's also another OS/disk Streaming product, which compares even more to old Ardence than netboot/i- Flex does.

It was first developed by a French company named Qualystem Technology, which Neoware acquired in 2005. Since Neoware was acquired by HP in 2007, now this product is available in HP portfolio.

It's name is HP Image Manager.

There is a free trial version:

h20392.www2.hp.com/.../displayProductInfo.do

Free for 90 days and 20 clients.

If you want to order it as a stand-alone software product, it's internal SKU in HP's ordering system is AT488AAE

It's much cheaper than Citrix PVS or DoubleTake Flex and lighter too. It can run entirely with Linux based servers. It might not have all the bells and whistles but it is a very good product for streaming a virtual disk to any kind of desktop, including VMs.

Help and support can be obtained through HP forums.

The most accurate forum is "remote PC solutions":

forums11.itrc.hp.com/.../categoryhome.do

Make sure that the words "Image Manager" are in the subject of your post on this forum, then Image Manager specialists in HP are warned almost immediately and you usually would have an answer in the next business day.

Have fun!

Stealth mode off...

That licensing decisions is stupid, I had to re read it a few times... I didn’t want to believe it...

I am very close to both of these products as I wish to move fwd with a streamed fat client solution for labs in my organisation (1000 units), yes I work in the education sector.

I have run up both products lately with streamed Win 7 Ent X64 with App-V client in stream mode using local disk cache for both OS write, and App-V global cache. This works well as the App-V global pathing to local means only the first client cops the FB1 load.

I had the same concerns that Brian mentioned when you think ISCSI (TCP) vs UPD but im not so sure it really matters, esp after the initial boot load (I have 1 gig links to the endpoints).

One thing I’m trying to get out of both vendors is scaling.

Citrix doco is old ... and doesn’t show X64.

Im interesting in seeing X64 based streamed servers with: Nehalem CPUs, 24 GB+ of ram and 10 GbE Nics.

Citrix have done well in supplying documentation on architecture namely around high availability which is vital. Im still trying to hunt down similar doco from DT.

Thanks Jeff for posting your experiences, I assume you achieve part of your HA via having 2 SANS?

Did I mention that the citrix bundle licence approach is stupid. Education customers won’t sign up for Hosted VDI when they don’t want it or need it. AppDetective get your horse we are forming a posse...

You can name you horse -> Broker :-P

@rahvintzu  - yes we started with a 2 SAN HA cluster of Lefthand SANS. So picture a server environment where you might have 4 'nodes' (diskless servers, dual quad core etc) and your SAN cluster. If you run 3 nodes (say Web, App & DB) your 4th can be switched off (like on the power socket off). If either of your three other nodes were ever to fail, you just power up the 'spare' to the SAN volume occupied by the failed node. The same goes for performing upgrades / hardware work on nodes.

The other really neat thing about this sort of setup is that - say you have 5 nodes total, and two of these nodes are in an HA MS DB Cluster - rather than going into the DC and fixing the problem server there and then (with all the stress of working on a degraded DB cluster) you can again power up your spare to take your cluster back up to 'normal' remotely. So when you finally go in to fix the problem server (at a more sensible hour) - this then becomes your spare.

How many times have I seen people go in to 'fix' nodes in degraded clusters, and sods law dictates that this is a 'perfect' opportunity for a 'whoops' to take the other one node down!!!!

We are now very interested to look at booting HyperV via PXE direct to Flex  The HyperV server itself is running Flex, so by using a product like DoubleTake Backup / Availability - you could seamlessly DR / replicate your entire setup somewhere else - in real time! If you also add booting your clients off this system into the mix - this is basically an entire company DR solution!

So in the event of a scud missile strike, all your servers and all your clients could be booting up from a facility someplace else - in about 10 mins - roughly the recovery time of DT High Availability - with no data loss.

That's pretty cool stuff.

Jeff

Thanks for the extra info it was a good read.

My mention of HA was more in the streaming fabric/components.

Sounds like you are using SANs so you have redundancy built into the SAN HW. I was wondering more about the Flex Storage Server option and how the client handles fail over live between storage nodes.

Might have to check the user guide again or ask for more info.

@rahvintzu - Ah so as I understand you have two Flex storage servers each with their each with their own storage, one goes down and you want to ensure that clients can fail over to the other - in which case you would need data replicated between the two?

Or are you talking about another type of failure - such as a switch or network adapter?

@rahvintzu. This won't impact me given my license, but I hear your concern. I wonder how much money Citrix made from PVS standalone for them to really care. This is a bit like App-V in MDOP when I only care about App-V. I think Citrix is taking the view go for broad market of Desktop not just one model. Now if there was enough $$$ to justify a standalone product I am sure that would change their mind. But given you are from education I doubt you have the buying power. I am curious to understand if the new campus pricing thing they announced would solve your problem?

@ Jeff

Lets say u have data rep going on the back end.

I was thinking more about the ability of the end client to cut between storage servers automatically during session, without reboot or loss of data.

This is important if you are using servers as storage devices as they need to be patched and rebooted each month.

@ appdetective

I have a quote for PVS standalone, but im unsure when Xendestop 4 will cut in. I would say the citrix campus agreement could make things more expensive if they try to base on student count. This is the highest count for education more than staff more than devices.So the unit cost has to low to say the least.

Re MDOP we have a MS campus agreement, Currently MDOP is 6 times cheaper than other App Virtualisation solutions.

@Jeff

I think client failover requires a boot or reboot.

So it can select another target that has been assigned, need to ask the vendor to confirm.

From User guide:

The target is now associated with the Flex Client or Flex Client group. If multiple targets are associated, failover mode is selected by default. If the first iSCSI target is not available, the Flex Client or Flex Client group attempts to connect to and boot from each subsequent target until successful.

@ rahvintzu

Right - I think that currently on the 'default' standard setup targets that are assigned as 'failover' would involve a client reboot if the active target went down.

However - there are some configurations which would actually enable clients to 'survive' a target loss - but I think Steve is the best person to talk about those!

Also one other thing - 'traditional' SANs are generally designed for handling lower numbers of iSCSI connections - whereas Flex is designed to handle (I believe) 1,000s. So there is an valid place for Flex to be installed 'on top' of a system that itself was running on your more traditional SAN.  In this role, the underlying SAN would be clustered, perform other stuff such as volume snapshots - with your 'thinner' flex server would providing the client access.

Thank Jeff, for the real world advice :-)

Steve, could you please PM me.... so i can take this discussion offline.

Cheers

@rahvintzu, my email address is available from my contact info.

We have internal facing HA documents that I can share with you, and are in the process of making them available on the web.

How the client handles failover (post boot) - whether to our own or other iSCSI storage - can be achieved via multi-path failover on the iSCSI initiator (assuming failures in connectivity on distinct separate paths to your storage)

On the Storage Server side, we can provide HA on that along with our GeoCluster product - it is a tested solution for which we are preparing some documentation for.

Feel free to contact me directly on any other questions that might be better addressed offline.

tftp boot (without the streaming) has been around for a really long time.  I bought a slew of diskless Sun 50's in maybe 1984 or 85 for my developers.  We never had an issue - and that was over the old shared thick yellow ethernet cables that you had to tap at specified distances.  One cable for the whole building with collision detect and all.  Of course the boot images were a wee bit smaller back then!

@rahvintzu,

We've made the HA documents publicly viewable here:

www.doubletake.com/.../Flex_v3.1_HA_for_Flex_Server_failover.pdf

www.doubletake.com/.../Flex_V3.1_HA_for_NIC_failover-postboot.pdf

www.doubletake.com/.../Flex_v3.1_HA_for_NIC_failover-preboot.pdf

Not a lot of text surrounding them on the PDFs, they are usually used  with in-person discussions on HA. They should convey enough information, otherwise, I hope!

@Steve

One of my colleague tells me that if an iSCSI client is disconnected from the network long enough, it looses "everything" and cannot resume iSCSI I/Os when reconnected. IOW, he said that if I disconnect for 5 minutes the network cable of a Flex client that booted of an iSCSI target, then I have to reboot the client after having connected the cable again for it to reconnect to the iSCSI target.

Is it true?

Thanks in advance

- Yves

@Yves

Long time since we last talked!

We have tested pulling a cable before - and it survived. I have also often suspended virtual machines booted via sofitware iSCSI initiator  (via Flex) and they have resumed fine as well....days later.

I cannot at this point state how consistent that behaviour is.  It may depend on the type of I/O that is running and how dependent any such application is on that. There is always the option to use multi-path failover with a second NIC.