Written on 10:27 AM by Right Click IT - Technology Services
– Jeff Aaron, VP Marketing, Silver Peak Systems, CIO November 18, 2008
IT departments can reap enormous benefits from virtualizing applications and implementing Virtual Desktop Infrastructures (VDI). However, the management and cost savings of virtualization can be lost if performance is so bad that it hampers productivity, as can happen when virtual applications and desktops are delivered across a Wide Area Network (WAN).
For an in-depth look at a WAN revamp, see CIO.com's related article, "How to Make Your WAN a Fast Lane: One Company's Story."
How can enterprises overcome poor performance to reap the rewards of virtualization?
Jeff Aaron, VP of marketing at Silver Peak Systems, suggests these five tips.
1. Understand The Network Issues
For starters, it makes sense to understand why your virtualized applications and virtual desktops perform poorly across the WAN. It's typically not due to the application or VDI components, but due to the network. More specifically, virtualized environments are sensitive to the following WAN characteristics:
Latency: the time it takes for data to travel from one location to one another.
Packet loss: when packets get dropped or delivered out of order due to network congestion they must be re-transmitted across the WAN. This can turn a 200 millisecond roundtrip into one second. To end users, the virtual application or desktop seems unresponsive when packets are being re-transmitted. They start to re-hit the keys on their client machines, which compounds the problem.
Bandwidth: WAN bandwidth may or may not be an issue depending on the type of traffic being sent. While most virtualized applications are fairly efficient when it comes to bandwidth consumption, some activities (such as file transfers and print jobs) consume significant bandwidth, which can present a performance challenge.
2. Examine WAN Optimization Techniques
WAN optimization devices can be deployed on both ends of a WAN link to improve the performance of all enterprise applications. The following WAN optimization techniques are used by these devices to improve the performance of virtual applications and desktops:
Latency can be overcome by mitigating the "chattiness" of TCP, the transport protocol used to by virtual applications for communication across the WAN. More specifically, WAN optimization devices can be configured to send more data within specific windows, and minimize the number of back and forth acknowledgements required prior to sending data. This improves the responsiveness of keystrokes in a virtual environment.
Loss can be mitigated by rebuilding dropped packets on the far end of a WAN link, and re-sequencing packets that are delivered out of order in real-time. This eliminates the need to re-transmit packets every time they are dropped or delivered out-of-order. By avoiding re-transmissions, virtual applications and desktops appear much more responsive across the WAN.
Bandwidth can be reduced using WAN deduplication. By monitoring all data sent across the WAN, repetitive information can be detected and delivered locally rather than resent across the network. This significantly improves bandwidth utilization in some (but not all) virtualized environments.
3. Set Application Priorities
The average enterprise has more than 80 applications that are accessed across the WAN. That means that critical applications, including terminal services and VDI, are vying for the same resources as less important traffic, such as Internet browsing. Because virtual applications and desktops are sensitive to latency, it often makes sense to prioritize this traffic over other applications using Quality of Service (QoS) techniques. In addition, QoS can guarantee bandwidth for VDI and virtual applications.
4. Compress and Encrypt in the Right Place
Often times host machines compress information prior to transmission. This is meant to improve bandwidth utilization in a virtual environment. However, compression obfuscates visibility into the actual data, which makes it difficult for downstream WAN optimization devices to provide their full value. Therefore, it may be a better choice to turn off compression functionality in the virtual host (if possible), and instead enable it in the WAN optimization device.
Moving compression into the WAN optimization device has another added benefit: it frees up CPU cycles within the host machine. This can lead to better performance and scalability throughout a virtual environment.
IT staff should also consider where encryption takes place in a virtual infrastructure, since encryption also consumes CPU cycles in the host.
5. Go With the Flows
Network scalability can have an important impact on the performance of virtual applications and VDI. The average thin client machine has 10 to15 TCP flows open at any given time. If thousands of clients are accessing host machines in the same centralized facility, that location must be equipped to handle tens of thousands of simultaneous sessions.
When it comes to supporting large numbers of flows, there are two "best practice" recommendations. First, as discussed above, it is recommended that compression and encryption be moved off the host machine to free up CPU cycles. Second, make sure your WAN acceleration device supports the right amount of flows for your environment. The last thing you want to do is create an artificial bottleneck within the very devices deployed to remove your WAN's bottlenecks.
Posted in
Network scalability,
Quality of Service (QoS),
RClick,
RClick.com,
Right Click,
Virtual Desktop Infrastructures (VDI),
virtualizing applications,
WAN optimization
|
Written on 10:33 AM by Right Click IT - Technology Services
– Julie Bort, Network World May 07, 2007
No. 1: Fine-tune your IPS.
"There's a lot of set-it-and-forget-it mentality in intrusion-prevention system marketing, and it's dangerous," says David Newman, president of testing facility Network Test and a Network World Lab Alliance member.
Fuzzing, in which the exploit is changed just enough for the security mechanism to miss it, trips up many IPSs, Network World's recent IPS test showed.
Network managers need to understand how each exploit works and how their IPS detects them, and then upgrade that protection routinely.
No. 2: Sell security by its benefits.
Start selling security to the purse-holders the way you do all other technology investments -- in measurable terms that relate to the business, recommends Mandy Andress, president of testing facility ArcSec Technologies and a Network World Lab Alliance member. Rather than saying how dangerous viruses are as a method to gain the budget for a reputation services antispam defense, for example, illustrate how much productivity could be gained by adding another layer of antispam control.
No. 3: Automate desktop and network access.
Wireless badges can come in handy for automated access control to desktop PCs, particularly those shared by multiple users in medical exam rooms, warehouses, call centers and the like.
For example, Northwestern Memorial Physicians Group implemented Ensure Technologies' XyLoc MD, which uses 900MHz radio-frequency technology encoded on staff ID badges for authentication, says Guy Fuller, IT manager at the Chicago healthcare organization. This saves the staff time while ensuring that network access and sensitive information are not available to other users.
No. 4: Link physical access to enterprise applications.
IP-based building-access systems built on industry-standard servers and using the existing data network are more affordable than ever because of open architecture products. Advances in server-management technology mean these systems not only are deployable by network (rather than the physical security) staff but are centrally manageable. Plus, they can integrate with ERP applications and network access-control systems.
Georgia-Pacific, a US$20 billion paper manufacturer in Atlanta, is rolling out Automated Management Technologies' WebBrix, an IP-based building-access system, to the majority of its 400 locations. IT used WebBrix's open application interface to write a custom application called Mysecurity that integrates the system with SAP, among other duties. When employees swipe their badges to gain access to the building, they also are sending data to SAP for time and attendance tracking, says Steven Mobley, senior systems analyst at Georgia-Pacific.
No. 5: Delegate an operating systems guru.
"Operating systems configuration can seem to some like a black art," says Tom Henderson, principle researcher for testing facility ExtremeLabs and a Network World Lab Alliance member. Setting the wrong combination is bad news. For example, large memory-block move options can affect the amount of dirty cache with which the operating system must deal, he says. If memory/caching options are balanced incorrectly, the machine could freeze. By assigning a staffer to master the voluminous documentation published by mainstream operating system vendors, servers can be safely fine-tuned to optimal performance for every application. The guru also should master Web server and BIOS setting options.
No. 6: Use VMware server memory smartly.
Without spending a dime, you may be able to boost the amount of memory available on virtualized Windows 2003 physical servers, thereby improving performance of the virtual machines. If all the virtual machines on the same physical box need the same memory-resident code, such as a dynamic link library (DLL), you can load the DLL once into the physical server's main memory and share that DLL with all virtual machines, says Wendy Cebula, COO at VistaPrint, an international online printer with U.S. operations headquartered in Lexington, Mass. "We've gotten big memory usage benefits by caching once per physical box rather than once per usage," she says.
No. 7: Move applications to a Linux grid.
If you have compute-intensive mainframe applications, don't shy away from lower-cost alternatives such as grid computing because the applications were written in COBOL, says Brian Cucci, manager of the Advanced Technology Group at Atlanta-based UPS, which has such a grid. The application will likely have to be redesigned somewhat for the new hardware platform. But vendors can be counted on to help, as they'll want to ally on the new technology.
No. 8: Recognize WAN links may degrade VoIP QoS.
This is particularly true in areas of the country where the public infrastructure is aging, says Bruce Bartolf, principal and CTO of architecture firm Gensler, in San Francisco. Having completed VoIP installation at seven of 35 sites, Bartolf found unexpectedly high error rates or complete failure on many links. To provide the kind of uptime and quality demanded of phone service, you need to design with alternative failover paths on the WAN. Cable may not be much better, but Metro Ethernet, if available, could work well, he says.
No. 9: Ease IP management with an appliance.
Although the tasks that appliances perform can be done with each vendor's gear, "with something as important as IP management, if you don't do it well, you can really hurt your five-nines," Gensler's Bartolf says. He chose Infoblox appliances, which manage numerous tasks, including Trivial File Transfer Protocol (TFTP) firmware upgrades. "Rather than dealing with Microsoft distributed file system, loading a TFTP server on a Microsoft server, running DHCP on a Microsoft server, running SMS on top of that, and managing it all, I have an appliance," he says. "I put it in, and it works."
No. 10: Shelve the fancy visuals.
"We found it highly impractical to make our monitoring visual," VistaPrint's Cebula says. VistaPrint relies on remote monitoring to manage its data centers, including one in Bermuda. It uses homegrown tools to track everything from CPU usage to event correlation. Visual graphing of events slowed down detection and analysis, taking network operations staff an average of five to seven minutes per event to use, Cebula says. When the tools used simple red, yellow and green lights, detection and correlation dropped to one or two minutes per event, she says.
And don't forget to keep your monitoring tools on at all times and run spot checks, advises Barry Nance, independent consultant and Network Lab Alliance member. The most common mistake is not to turn them on until an event occurs.
Posted in
automate desktop,
enterprise applications,
IPS,
IT security,
Linux,
network access,
operating systems guru,
QoS,
RClick.com,
Right Click,
technology tips,
VMWare certified,
VoIP,
WAN optimization
|
