Autocache, is an I/O acceleration product which uses flash devices. It is installed on the hypervizor, inspects the I/O from all virtual machines and caches the hot I/O onto a local PCIe flash card or solid-state disk (SSD). It is a read accelerator, which means that writes go directly to persistent storage (write-thru), but those writes are cached so that future read operations are serviced from flash. Of course this is also beneficial for writes, since if the reads are serviced by flash, the spinning disks are not accessed for reads, thus more writes. Autocache works for virtual machines on both block and NFS.
Since it is installed on the hypervizor, Autocache does not install any agents in the guest OS, It supports vSphere HA and core vSphere features such as vMotion. It is integrated into the existing workflows in the VMware vSphere client and is managed by an Autocache tab in the client. Here is an example of a statistics view from the Autocache tab in the vSphere client:
The fact that this is an agent-less product, fully integrated with core vSphere features like HA & DRS, supports vMotion and has negligible overhead on the hypervizor makes this a very interesting product. Proximal data claim negligible CPU overhead and memory consumption of about 1/1400th of flash capacity. Proximal Data tell me that customers can typically expect a 2x-3x in virtual machine density, depending on workload.
Currently, Autocache supports VMware ESXi 4.1 and 5.0/5.1.
I’ll admit that I haven’t had the opportunity to try out the product – this time of year is always rather busy for us in VMware Technical Marketing. The content here is based on conversations with Rich over the last couple of months. However, Proximal Data allow you to download and evaluate the product, so I’d say give this a go for yourselves. The download is available here. If you do get a chance to evaluate, I’d love to hear your thoughts so please leave a comment.
[Update] Proximal Data were acquired by Samsung in November, 2014. More here.