File Handling - Its Impact on Server Backups
By: Chris Greer
Server backups. The silent source of frustration in certain data centers. Whether they are simply firing at times when production traffic is contending for server resource, or are just taking hours longer than they should, backups can cause headaches for network managers. In this article we are going to take a look at how file handling can cause a backup to run much longer than it should, and how it contributes to lower overall throughput.
In any backup process, regardless of what software or method is used, the goal is to move data from one location to another as quickly as possible. Many things can cause the backup to slow to a crawl, and oftentimes they are completely unrelated to network bandwidth. One example of a throughput killer is file handling – when individual files or directories are requested by the backup server. These requests require the source server to call the file, queue it for delivery then transmit the file. If the server delays at all in its response to these individual file requests, this can significantly reduce the overall throughput utilized by the backup, especially when several thousand files are handled in this way.
In one example, the backup software was reporting that the backup process was transferring at less than 30 Mbps, when both servers and all network connections between were connected at 1Gbps. Overall network utilization was very low, measuring at only 5%.
For some reason, the backup process was not taking advantage of these high-speed links.
Using Wireshark, a capture was taken during the slow backup. We first looked for retransmissions in the traffic between the two servers. This would indicate packet loss and would direct us to look at the switches for Ethernet errors such as FCS and Alignment Errors. The traces showed no retransmissions, only slow response times from the source server to file requests. Every file was requested individually and would incur a 20 millisecond delay on the source server. When this file handling was filtered out of the trace, backup throughput measured at almost 1Gbps, fully utilizing the available bandwidth.
The chart below shows the impact of file handling on data backups. This data was compiled assuming the total data is 100GB. When the backup begins, only a small amount of this data will be requested at once, which depends on the protocol in use. Typically, a SMB/CIFS based backup application will read blocks of data as shown below. The smaller the blocks requested, the more are required to move the total data amount. This will incur more read delays, which reduce overall throughput.
As shown above, file handling can have a tremendous impact on bandwidth used during file backups. In many cases, slow performance can be traced to slow file responses on the source server. Use of multi-threaded backup applications, monitoring on the source server for resource problems, or backup applications with less file handling overhead will improve backup performance in these cases. Performance can also be impacted by TCP Window issues as well as underlying network problems. Specific instances of these problems and their resolutions will be discussed in future articles.
About the Author:
Chris Greer is a Senior Network Analyst for Network Protocol Specialists, a Seattle based Network Consulting company. Chris has 10 years of experience in analyzing and troubleshooting networks. He regularly assists companies in tracking down the source of network and application performance problems using a variety of protocol analysis and monitoring tools including Wireshark. When he isn’t hunting down problems at the packet level, he can be found teaching various analysis workshops at Interop and other industry trade shows. Chris also delivers Fluke Networks public courses and protocol analysis themed webcasts. He can be contacted at chris (at) nps-llc (dot) com.
This Article is Brought to you by:
Article Sponsorships Available
Short description about your link.
Add your link here
Article Sponsorships Available
Short description about your link.
Add your link here
Network Analysis Tips - Analyzers - Packet Sniffer Related Articles:
Datacom Systems Inc. - SINGLEstream 1214LX-10G
Datacom Systems Inc. - SINGLEstream™ 1214LX-10G Aggregation The SINGLEstream™ 1214LX-10G can combine data from up to five single ...
SNMP Tools
AdventNet SNMP API can be used to build system management, application management and network management applications and applets. It includes class librarie...
By: SNMP Tools
Updated Ethernet Related News:
The all-Ethernet data center took a step forward this week with Xsigo's introduction of its Ethernet-based IOV solution at VMworld. Xsigo is hardly the first vendor to introduce I/O Virtualization sol...
Brocade adds 100G Ethernet to switch and router line
Super fast networking standard to enter products Brocade will soon unveil 100Gbps Ethernet additions to its switching and routing product line, in what the company claims is its most significant Ether...
Ethernet switch takes new tack at virtual I/O
Startup Xsigo Systems, Inc. rolled out a 10 Gbit/second switch that can handle Ethernet networking and Fibre Channel storage traffic for a fraction of the cost of Fibre Channel over Ethernet adapters ...
Brocade set to unveil 100G Ethernet
Brocade will soon unveil 100Gbps Ethernet additions to its switching and routing product line, in what the company claims is its most significant Ethernet/IP product rollout since acquiring Foundry Ne...
Xsigo Gives Ethernet a Path Into Converged I/O
Virtual I/O vendor Xsigo is introducing Ethernet-based versions of its I/O Director products. Ethernet - Xsigo - Data Communications - Manufacturers - Performing Arts
