This post contains some of the things I remembered from VMWorld Europe 2013.
- VMware is working on a tool to help with a sizing of VMWare Flash Read Cache size and block size.
- Using VSAN to build metro-cluster? Currently not since it’s not possible to control location of replica and witness. They all might end up in the same data center. In the future this might change when configurable fault-domains are introduced.
- SAP HANA performance is 95% of bare metal performance when running on VMWare
- When vNUMA is bigger then pNUMA performance is impacted.
- When increasing vCPU count change virtual sockets not virtual cores. Change virtual cores only if needed by operating system license.
- Rightsize your VMs! Idle vCPU-s still need management. Idle vCPUs will not increase application performance.
- Space reclamation has been improved in vSphere 5.5
- EMC ViPR will be available from January 2014.
- Pure Storage biggest setup can be 36TB raw. Performance specifications are for one storage controller – one controller can do 400K 8KB IOps.
- vSphere Flash Read Cache allocation is a hard reservation at the moment. This might change in the future.
A good understanding of the workload behavior and characteristics is the most important factor in deciding whether or not to enable vFRC because not all workloads will benefit from vFRC. vFRC caches data from both read and write I/Os, but write I/Os are always serviced by the underlying storage. Therefore, workloads that have a majority of reads can directly benefit from vFRC. Cache block size is the minimum granularity of cache fills and cache evictions. Having the optimal cache block size is critical to overall performance of vFRC. (Source: VMWare )
To set optimal cache block size we need to find a most frequent I/O size of the workload. VMWare ESXi has a tool named vscsistats for I/O size analyzing. Instead of vscsistats I have been using another free tool – EMC XtremSW Cache Performance Predictor.
EMC XtremSW Cache Performance Predictor main function is to generate prediction reports based on host traces. Cache Reads Analysis report predicts benefits in IO workloads when using host based EMC XtremSF PCI Flash cards. But it also contains more valuable information – time window of trace, amount of I/O’s recorded, Workload Mix chart and I/O Size Distribution chart.
Workload mix chart shows read to write ratio. Big read percentage increases the chance for performance benefits when using vFRC. I/O Size Distribution chart helps you to select the best vFRC cache block size.
I have not yet upgraded my VMWare environment to 5.5 but I have already found some candidates for vFRC. My suggestion is to start with database servers. They are most likely candidates for performance improvements.
EMC XtremSW Cache Performance Predictor tool can be downloaded from EMC Support site – link.
More information about EMC XtremSW Cache software – link and link.
HP has published a new firmware 126.96.36.199 for G7 and Gen8 blades.
Update resolves the following issues for CNA firmware (source: HP ):
- Unable to set speed/duplex to auto/auto and could not make Host profile
- Firmware memory leak with multiple logins to a redirected target
- HP NC553i 10Gb 2-port FlexFabric Converged Network Adapter locks up during iSCSI login process
- Lost network connectivity after firmware update
- VMware PSOD (Purple Screen of Death) on HP NC551i Dual Port FlexFabric 10Gb Network Adapter and HP NC551m Dual Port FlexFabric 10Gb Converged Network Adapter on AMD servers.
- HP NC553i 10Gb 2-port FlexFabric Converged Network Adapter lost LAN and SAN connectivity
Release notes can be accessed and ISO can be downloaded from here.
HP as also released online firmware update package for VMWare which can be downloaded from here.
I tried the online update package on BL460c G7 running ESXi 5.0. During the firmware update there was no disruption to the network or FC connectivity. After the update server must be restarted to activate the new firmware.