Syndicate this forum's content with:   RSS feed Add to My Yahoo! Add to Google

DataCore Software Community Forum

Accelerate AIM TransferRate

Share/Bookmark: Add this Thread to Digg Digg   Add this Thread to StumbleUpon Stumble Upon   Add this Thread to Del.icio.us Del.icio.us   Add this Thread to Reddit Reddit   Add this Thread to Facebook Facebook
RE: Accelerate AIM TransferRate
by BContario on Thu, May 6 2010 12:29 PM
The OpenVPN solution is very ingenious and cost-effective, however I am experiencing degraded throughput using OpenVPN compared to a straight-shot connection that I am hoping someone can either confirm is a showstopper or is fixable.

I am using OpenVPN 2.1.1 win32 on Windows 2003 R2 32-bit. Functionally it works great and handles compression just as described in earlier postings with no encryption (since we have an IPSEC VPN set up already). However, we seem to be hitting a ceiling where we can never seem to get more than 7-8Mbit average data throughput, which translates to roughly 3-4Mbit compressed throughput going through the firewall.

If I turn off compression but still use the OpenVPN connection the throughput drops to 3-4Mbit.

If I use the normal IP addresses and disable OpenVPN I get around 50Mbit throughput on our 100Mbit internet connection.

The last test I did was to set up public IP addresses for the AIM Source and Destination server and use OpenVPN to make the point to point connection bypassing our corporate VPN, but I still only got 7-8Mbit using compression.

To recap:
OpenVPN + Compression thru IPSEC = 7-8Mbit AIM, 3-4Mbit Internet
OpenVPN - Compression thru IPSEC = 3-4Mbit AIM, 3-4Mbit Internet
OpenVPN + Compression direct = 7-8Mbit AIM, 3-4Mbit Internet
NO OpenVPN thru IPSEC = 50Mbit AIM, 50Mbit Internet

So, it appears that OpenVPN has some kind of ceiling on the throughput it allows, even though I have plenty of CPU available. I know about the --shaper option for OpenVPN and have tried using it and not using it and it doesn't seem to make a difference. One possible issue is that OpenVPN creates a virtual NIC that registers itself as a 10Mbit NIC. I am wondering if even though elsewhere on the net there are postings saying that number is not important and they get 100+ Mbit throughput using OpenVPN, they might be using a Linux build, or there may be something in AIM or IIS FTP that inspects the link speed to try to do some preemptive flow control versus just letting it saturate the link in my instance.

Any thoughts on what I might try (other than a $10K+ WAN Accelerator appliance)? We commit to paying for 20Mbit of bandwidth on a 100Mbit internet connection, and overage charges are $100 per Mbit, so we want to limit our AIM usage of our internet connection to about 10Mbit. But were hoping to be stuffing 20-30Mbit of AIM data through that 10Mbit of compressed data, and 7-8Mbit using OpenVPN as it stands is just too slow.

Thoughts, advice, and fixes appreciated. Thank you in advance!
BContario
BContario
Joined: Sat, Nov 29 2008
Posts: 10
RE: Accelerate AIM TransferRate
by abnormaliti on Thu, Jan 28 2010 5:55 PM
It has been a while since i completed the trial.

Hardware acceleration was 6-8 times base rate compared to OpenVPN which is pretty consistently 3 times.

Since the trial i have been happily running w/ OpenVPN and even an upgrade to W2K8 x64 didn't cause issues.

However, with the steady increase in data being replicated i have scheduled the installation of Juniper WAN accelerator next month.
abnormaliti
abnormaliti
Joined: Wed, Mar 4 2009
Posts: 7
RE: Accelerate AIM TransferRate
by cjayd on Mon, Jan 25 2010 3:06 AM
Hi
How did you OpenVPN transfer rates compare to the hardware WAN Accelerator.

Kind Regards
cjayd
Joined: Sun, Jan 24 2010
Posts: 3
RE: Accelerate AIM TransferRate
by abnormaliti on Mon, Mar 9 2009 8:08 AM
WANscaler and Riverbed are all good ideas but you can not bet the price of this openvpn solution.

This week i have a few Juniper WAN accelerators arriving for a 1 month trial.

It will be interesting to see how much better the transfer rate is compared to a simple openvpn tunnel.
abnormaliti
abnormaliti
Joined: Wed, Mar 4 2009
Posts: 7
RE: Accelerate AIM TransferRate
by floydpepper on Mon, Mar 9 2009 7:51 AM
Plattenschubser Wrote:Hi, I can't open the link, is there something wrong?


corrected the link - please try again
"I wouldn't listen to the trash I wrote." - Sgt. Floyd Pepper
floydpepper
floydpepper
Joined: Fri, Oct 3 2008
Posts: 73
RE: Accelerate AIM TransferRate
by Plattenschubser on Mon, Mar 9 2009 7:25 AM
Hi,

I can't open the link, is there something wrong?
Plattenschubser
Plattenschubser
Joined: Thu, Feb 26 2009
Posts: 2
RE: Accelerate AIM TransferRate
by floydpepper on Thu, Mar 5 2009 1:19 PM
AIM traffic seems to be pretty good compressible. There's a whitepaper available on DataCore's website for optimizing WAN links with Riverbed appliances: Link: http://www.datacore.com/downloads/DataCore_Riverbed_DR_22Sep08.pdf
"I wouldn't listen to the trash I wrote." - Sgt. Floyd Pepper
floydpepper
floydpepper
Joined: Fri, Oct 3 2008
Posts: 73
RE: Accelerate AIM TransferRate
by SANpaul on Thu, Mar 5 2009 6:02 AM
You can get even better results using Citrix WANscaler. I´ve seen effective transfer rates of up to 15 MBit/s over a 2 MBit/s SDSL link when using a WANscaler to optimize AIM traffic.
SANpaul
SANpaul
Joined: Fri, Oct 17 2008
Location: Dortmund, Germany
Posts: 16
RE: Accelerate AIM TransferRate
by abnormaliti on Wed, Mar 4 2009 7:04 PM
Heres the spreadsheet
xlsx Datacore AIM tunnel concept.xlsx, 30.924k
abnormaliti
abnormaliti
Joined: Wed, Mar 4 2009
Posts: 7
Accelerate AIM TransferRate
by abnormaliti on Wed, Mar 4 2009 6:57 PM
Last Edit: Wed, Mar 4 2009 7:01 PM
After moving our DR SDS offsite I was disappointed in the AIM throughput that I was seeing through the WAN so I began investigating ways to improve that throughput.

My investigations lead me to WAN acceleration/optimisation solutions. These devices use several methods to improve throughput and responsiveness, an important component is compression. So it occurred to me that Inline compression could be easily achieved with a point-to-point tunnel.

I do a lot of work with site-to-site VPN?s and often see benefits from enabling compression so I decided to give an OpenVPN tunnel a go and the results are surprising.

Here is the information and I hope others may find it useful.

Proof-of-concept design:
-       dcs01 = primary SDS, AIM source
-       dcsdr01 = DR SDS, AIM destination
-       Existing and working AIM relationship using FTP as the transport method
-       OpenVPN 2.1 rc15. Site: www.openvpn.net
o       Encryption and security disabled.
o       Compression enabled.
o       dcs01 VPN IP: 192.168.199.2
o       dcsdr01 VPN IP: 192.168.199.1

Steps:
1.       Install OpenVPN on AIM source and destination servers. I used OpenVPN 2.1_rc15.
2.       For the purposes of testing the concept I initiated a tunnel from the command line.
a.       On dcs01.
i.       Command: openvpn --cipher none --comp-lzo yes --remote dcsdr01 --ifconfig 192.168.199.2 192.168.199.1 --verb 4 --dev tun
b.       On dcsdr01
i.       Command: openvpn --cipher none --comp-lzo yes --remote dcs01 --ifconfig 192.168.199.1 192.168.199.2 --verb 4 --dev tun
3.       On dcs01 using AIM source manager GUI change the IP Address of the dcsdr01 destination server to 192.168.199.1.
4.       Thats it.

Assuming there are no firewall rules in the way and IIS FTP service on dcsdr01 is not bound to a specific IP or restricting access to a specific IP, AIM data will start moving across the tunnel.

In summary average throughputs were (assuming the perfmon counter is KB/sec):

-       With OpenVPN: 2902
-       Without OpenVPN: 1304

Attached is a Excel of some perfmon data.

Notes:
-       To keep the AIM buffer full I did Initialize Volume on a couple of volumes and performed some zero fill on another (may as well kill two birds with one stone by reclaiming some SAU?s.). That is in addition to general data from all the production volumes.
-       The WAN is not dedicated for this purpose, i.e. other data was on the line during the test however the test was carried out after hours to keep other traffic to a minimum.
-       I am certain the type of data being replicated would be a factoring the compression, in this test the zero fill data would have resulted in excellent compression.
-       In my production environment AIM without the tunnel cannot keep up and by the end-of-business the time latency values from busy volumes have blown out to several hours however with the compression tunnel it is able to keep that as low as a few minutes.

Ben
abnormaliti
abnormaliti
Joined: Wed, Mar 4 2009
Posts: 7