I am using X2Go Client and we are facing Latency issues when developers use it from different country and they use graphics functionalities.

I'm new to x2go, but have a good bit of experience administering FreeNX for business desktops served over WANs. In Session Preferences -> Connection, experiment with the slider settings. In particular, MODEM, ISDN, and ADSL. I find that ADSL works best over my 3 - 10 mbit/s WANs. But if yours are slower, MODEM or ISDN might work better. On a fast WAN, MODEM and ISDN make the updates clunkier by limiting the update rate, and possibly other things. It's hard to predict all interactions. But you might even try 'WAN' for a relatively high bandwidth, high latency connection. The higher slider levels are good for latency. But note that 'LAN' provides no compression at all. It's likely to perform quite badly.

Also, on that same page, you can reduce the image quality, which defaults to 9 (highest quality) but supports a range from 0 to 9. I'm not familiar enough with the "Method" options to make a recommendation. I suspect the default is probably pretty good.

To give you an idea what you should be able to expect, NX technology over a 3mbit full duplex link with a ~150 - 200ms latency between sites provides very usable business desktops (Mail, browsing, Libreoffice) for my ~100 desktop users. About 50 of those are over the WAN, and the rest are on the local LAN. I know of nothing that even remotely compares to FreeNX/x2go for performance over a WAN.

Please experiment and report back. I'm interested in your results. And please tell us more about your problematic workload. The ping times to the clients, the nominal bandwidth, etc.

-Steve

Thanks Steve and Mike for the guidance you have provided. I really appreciate your help.

My servers are available in California and the developers are working in Europe and in India. So the ping time varies from 150 ms to 400 ms. I have selected ADSL with 16m-jpeg connection. The developers are accessing Linux environment with GNome. They use some Java 2D graphics for the development. Their normal bandwidth (tested using Speedtest.net) is around 3-5 MBPS.

The replies from Mike and Steve, I understood that for this kind of high latency I should use "WAN" connection speed with compression method - "No Pack". Is this my understanding correct?

I have already set Image quality to 1 and disabled audio/print/shared folder support.

I have also used libjpeg-turbo library on the server side (Linux OS).

Will this kind of setup work? If I upgrade hardware on my server/client, will it make any difference? Can X2go handle this kind of latency issues with some settings? Is there any minimum bandwidth required with this latency?

Thanks, Nirav

On Sat, Nov 23, 2013 at 4:24 PM, Mike Gabriel < mike.gabriel@das-netzwerkteam.de> wrote:

Hi all,

thanks Steve for your contribution. Very good.

On Sa 23 Nov 2013 20:11:25 CET, Steve Bergman wrote:

I am using X2Go Client and we are facing Latency issues when

...

...

developers use it from different country and they use graphics functionalities.

I'm new to x2go, but have a good bit of experience administering FreeNX for business desktops served over WANs. In Session Preferences -> Connection, experiment with the slider settings. In particular, MODEM, ISDN, and ADSL. I find that ADSL works best over my 3 - 10 mbit/s WANs. But if yours are slower, MODEM or ISDN might work better. On a fast WAN, MODEM and ISDN make the updates clunkier by limiting the update rate, and possibly other things. It's hard to predict all interactions. But you might even try 'WAN' for a relatively high bandwidth, high latency connection. The higher slider levels are good for latency. But note that 'LAN' provides no compression at all. It's likely to perform quite badly.

Also, on that same page, you can reduce the image quality, which defaults to 9 (highest quality) but supports a range from 0 to 9. I'm not familiar enough with the "Method" options to make a recommendation. I suspect the default is probably pretty good.

To give you an idea what you should be able to expect, NX technology over a 3mbit full duplex link with a ~150 - 200ms latency between sites provides very usable business desktops (Mail, browsing, Libreoffice) for my ~100 desktop users. About 50 of those are over the WAN, and the rest are on the local LAN. I know of nothing that even remotely compares to FreeNX/x2go for performance over a WAN.

Please experiment and report back. I'm interested in your results. And please tell us more about your problematic workload. The ping times to the clients, the nominal bandwidth, etc.

With connection speed X2Go (i.e. NX) is very good, even on very low speed lines.

However, with latency issues things become different. As latency is defined as the amount of time the data needs to traverse from client to server / server to client, we should make sure that on high latency connections, data is as processed as fast as possible.

So my basic idea about this is, that what you actually need to reduce is the time that is required for processing/caching/compressing images etc. On high latency lines it helps to have good bandwidth available, because you (in theory) can use the high bandwidth to compensate for high latency. E.g. by choosing a faster but less effective algorithm for compression (or not compressing images at all).

My personal problem here: I cannot test this theory, because I do not have high latency connections here to test this with. I know that you can simulate high latency / low bandwidth in a lab setup, but I neither have time nor resources for doing that.

Greets, Mike

--

DAS-NETZWERKTEAM mike gabriel, herweg 7, 24357 fleckeby fon: +49 (1520) 1976 148

GnuPG Key ID 0x25771B31 mail: mike.gabriel@das-netzwerkteam.de, http://das-netzwerkteam.de

freeBusy: https://mail.das-netzwerkteam.de/freebusy/m.gabriel%40das- netzwerkteam.de.xfb

---

X2Go-User mailing list X2Go-User@lists.berlios.de https://lists.berlios.de/mailman/listinfo/x2go-user

-- Thanks,

Nirav Shah (C) (412) 296-9491

Thanks Mike for the quick and helpful answers..See my updates in bold.

On Sat, Nov 23, 2013 at 5:35 PM, Mike Gabriel < mike.gabriel@das-netzwerkteam.de> wrote:

Hi Nirav,

On Sa 23 Nov 2013 22:47:23 CET, Nirav Shah wrote:

Thanks Steve and Mike for the guidance you have provided. I really

...

appreciate your help.

My servers are available in California and the developers are working in Europe and in India. So the ping time varies from 150 ms to 400 ms. I have selected ADSL with 16m-jpeg connection. The developers are accessing Linux environment with GNome. They use some Java 2D graphics for the development. Their normal bandwidth (tested using Speedtest.net) is around 3-5 MBPS.

The replies from Mike and Steve, I understood that for this kind of high latency I should use "WAN" connection speed with compression method - "No Pack". Is this my understanding correct?

I am really not an expert on this, but you could try that (maybe not ,,No Pack''). On IRC there is a guy called "TheUser". Ask him about that question once he appears again. He had another WAN based recommendation that I forgot again.

Using speed = LAN is identical to normal X11 over SSH.

• I will try to use LAN option and see if it is able to resolve the problem or not. I have also downloaded few tools like Charles and NLC tool for Mac to test the latency using my computer. I will try different option to solve this problem.*

I have already set Image quality to 1 and disabled audio/print/shared

...

folder support.

I have also used libjpeg-turbo library on the server side (Linux OS).

You should have libjpeg-turbo on the client, as well. But for this you need an nxproxy that uses the compat libs from libjpeg-turbo. If you build statically, then you need to build against libjpeg-turbo directly.

• I am not sure, how can I install this library on my client on windows and Mac. Is there any documentation available somewhere? as I am not aware of this library. *

Will this kind of setup work? If I upgrade hardware on my server/client,

...

will it make any difference? Can X2go handle this kind of latency issues with some settings? Is there any minimum bandwidth required with this latency?

All good questions that I do not have an experienced answer for. I only know that several other people have struggled with this.

Mike

DAS-NETZWERKTEAM mike gabriel, herweg 7, 24357 fleckeby fon: +49 (1520) 1976 148

GnuPG Key ID 0x25771B31 mail: mike.gabriel@das-netzwerkteam.de, http://das-netzwerkteam.de

freeBusy: https://mail.das-netzwerkteam.de/freebusy/m.gabriel%40das- netzwerkteam.de.xfb

-- Thanks,

Nirav Shah (C) (412) 296-9491

Java graphics are not that complex but it has run time debugging features which is performing(or displaying) very slow when we use x2go client. Is it possible to use Nxclient with X2go server? I am fine with that too. I have tried to use NXServer but I found it difficult to install/configure compared to X2go server.

I am going to try all the options if I am able to test different latency on my machine. I tried with Charles but somehow, it was not giving me the proper result of latency after configuring it.

On Sat, Nov 23, 2013 at 6:12 PM, Steve Bergman <sbergman27@gmail.com> wrote:

It's hard to get around the cold hard fact that a 400ms round trip time means that nothing that requires talking to the server can happen in less than 400ms, which is nearly half a second. (Consider that Einstein, himself, is imposing at least 50ms of that latency upon you from the grave.)

How large/complex are these java graphics? The NoMachine NX client has controls for both RAM and on-disk caches. The RAM cache is configurable from 4MB to 128M. And the on-disk cache is configurable from 0MB to 512MB. Depending upon the exact workload, large caches may help. But I'm not sure how to set that in x2go. (What does the '16m' in '16m-jpeg' actually mean, I wonder? 16m RAM cache? Something else?)

Regarding trying out 'LAN', I can tell you right now that's going to perform *extremely* badly. I'd have a lot more hope for 'WAN'. My best *guess* is that ADSL *may* be your best bet. But do try everything and see what works best.

-Steve

-- Thanks,

Nirav Shah (C) (412) 296-9491

Thanks Steve.

I have tried VNCs before and I found X2Go better than any other. I tried to throttle using Charles with specifying 300-400 ms latency but from my location, it always shows the same result (70ms ping time).

I am going to try some other tool to check this issue. It is possible to change few settings in java application like heap memory etc.

On Sat, Nov 23, 2013 at 7:51 PM, Steve Bergman <sbergman27@gmail.com> wrote:

No. Although both use the same underlying NX libraries, you can't mix and match X2GO/FreeNX clients/servers. (Although you *can* run x2goserver and FreeNX server in parallel on the same server. In fact, I'm doing this right now in preparation for our transition to x2go.) I'm finding x2go to perform at least as well as NX. And FreeNX is more or less an abandoned project at this point, as is NeatX. It's probably not worth the trouble of trying to get FreeNX installed and working on CentOS. (CentOS 4 was the last release that I've installed FreeNX on.) I've no reason to think FreeNX would do any better than x2go. But if there's a way to up the cache size in x2go, that might be worth a try. I've looked in the sessions and settings files in ~/.x2goclient/ and there don't seem to be any hidden options there that would apply. I'm guessing that we're probably getting the NX default values of 16MB RAM cache and 32MB disk cache, which ought to cover most situations reasonably well.

One other option that would be easy enough to try would be one of the VNCs. In general, x2go way outperforms them. But you never know. For this particular workload, something like tightvnc, tigervnc, or vnc4 might work better. I doubt it. But vnc is easy to set up for a test.

Any possibility that changing settings in the java application might mitigate the problem?

BTW, thanks for the reference to 'charles':

http://www.charlesproxy.com/documentation/proxying/throttling/

I didn't know about this, and had been looking for something similar.

-Steve

-- Thanks,

Nirav Shah (C) (412) 296-9491

Thanks Daniel. I tried to see WANem but it is not available for Mac but I am able to use Network Link Conditioner on Mac and able to get the expected latency.

Now, major issue is what are the best configuration for high latency setup.

On Sun, Nov 24, 2013 at 1:08 AM, Daniel Lindgren <bd.dali@gmail.com> wrote:

If anyone needs to set up a test environment for latency issues I can recommend WANem, "The Wide Area Network emulator". I've used it to emulate real world high latency links in my job environment and the results are very close to reality. Just remember to set half the latency on ingoing and the other half on outgoing packets, e.g. 100 + 100 to emulate a 200ms latency.

Cheers, Daniel

2013/11/24 Nirav Shah <shah.niravk@gmail.com>

...

Thanks Steve.

I have tried VNCs before and I found X2Go better than any other. I tried to throttle using Charles with specifying 300-400 ms latency but from my location, it always shows the same result (70ms ping time).

I am going to try some other tool to check this issue. It is possible to change few settings in java application like heap memory etc.

On Sat, Nov 23, 2013 at 7:51 PM, Steve Bergman <sbergman27@gmail.com>wrote:

...

No. Although both use the same underlying NX libraries, you can't mix and match X2GO/FreeNX clients/servers. (Although you *can* run x2goserver and FreeNX server in parallel on the same server. In fact, I'm doing this right now in preparation for our transition to x2go.) I'm finding x2go to perform at least as well as NX. And FreeNX is more or less an abandoned project at this point, as is NeatX. It's probably not worth the trouble of trying to get FreeNX installed and working on CentOS. (CentOS 4 was the last release that I've installed FreeNX on.) I've no reason to think FreeNX would do any better than x2go. But if there's a way to up the cache size in x2go, that might be worth a try. I've looked in the sessions and settings files in ~/.x2goclient/ and there don't seem to be any hidden options there that would apply. I'm guessing that we're probably getting the NX default values of 16MB RAM cache and 32MB disk cache, which ought to cover most situations reasonably well.

One other option that would be easy enough to try would be one of the VNCs. In general, x2go way outperforms them. But you never know. For this particular workload, something like tightvnc, tigervnc, or vnc4 might work better. I doubt it. But vnc is easy to set up for a test.

Any possibility that changing settings in the java application might mitigate the problem?

BTW, thanks for the reference to 'charles':

http://www.charlesproxy.com/documentation/proxying/throttling/

I didn't know about this, and had been looking for something similar.

-Steve

-- Thanks,

Nirav Shah (C) (412) 296-9491

---

X2Go-User mailing list X2Go-User@lists.berlios.de https://lists.berlios.de/mailman/listinfo/x2go-user

---

X2Go-User mailing list X2Go-User@lists.berlios.de https://lists.berlios.de/mailman/listinfo/x2go-user

-- Thanks,

Nirav Shah (C) (412) 296-9491