HP ProLiant DL380G5-WSS 3.7.0 HP StorageWorks HP Scalable NAS File Serving Sof - Page 251
usage is important if replication is to be successful., Replication increases the load on the CPU
View all HP ProLiant DL380G5-WSS manuals
Add to My Manuals
Save this manual to your list of manuals |
Page 251 highlights
• Replication increases the use of physical memory and swap space. If your system is running well with 8GB of main memory and you add replication to it at near the configuration limits, there is a substantial chance that the system will be unable to service its normal workload in addition to the replication workload. • Replication must have sufficient bandwidth between sites to allow data transfers to complete in a timely fashion. Calculating how much data flows between sites is difficult, since it is based upon the speed with which data changes, the replication interval, and many other factors. A rough guess for the bandwidth required is the amount of data that is changing per interval divided by the interval length. (The replication overhead in packet headers is made up for by the potential compression of data being transferred in this estimate). If your replicated load is composed primarily of small files or small changes, the bandwidth consumed by replication may be much lower than the full bandwidth available to it, due to latency in file access during the replication process. This can slow replication significantly. As a rule of thumb, during instance zero replication (when a new filesystem is being populated on the destination side of the link), when replication can be at its most efficient, the maximum replication rate for a 1Gb link is roughly 100-110 MB/sec. This means that a 10TB filesystem will take at least 27 hours to fully replicate. This is without any delays, errors, or network trouble, and using the replication network as hard as it is possible to use it. A non-instance zero case, where the changes are scattered throughout the filesystem, will use the network less effectively and thus will take longer to replicate a given amount of data. It easy to see that planning for replication bandwidth usage is important if replication is to be successful. • Replication increases the load on the CPU, especially on the replication sentinel node. • Replication loads the administrative filesystem heavily. System capacity must be increased for replication to proceed successfully at high loads. The system load is greatest is when there is a sudden surge of file or filesystem changes. It is difficult to predict the effects on the system of surges from the steady state performance of the system. For this reason, HP recommends that the system be over-configured for day-to-day use so that surges can be accommodated without problems. Replication has been tested with up to 16 nodes and 200 replication mount points. Clusters that are configured to these limits should expect that the nodes on which the sentinel is expected to run should be configured per the recommendations below. HP Scalable NAS File Serving Software administration guide 251