I’ve been invited as a delegate to Storage Field Day, in its 4th edition in San Josè, California. I met Avere Systems and I learned about their solution; our host has been their CEO and founder, Ron Bianchini.
Who is Avere Systems
Avere original idea has been to optimize NAS by eliminating performances problems and latency using their design, called Edge-Core. This solution can optimiza NAS, by sitting in between users and the Filer in the backend.
The solution, called FXT, is not a substitute of a NAS, it does not offer additional capacity, but acts as an optimizer, both for read and most of all write operations, but that’s not all. I/O acceleration alone is something even a simple caching system can do, but Avere also helps in directory browsing for example: this is a huge problem that can really slow down large NAS installations more than mere IOPS; Avere claims it can improve performances in this scenario up to 50 times. The FXT architecture can be increased up to 50 nodes, giving an impressive overall CPU and memory value, while on the contrary a NAS is usually sized upfront and it does not scale when you add shelves or disks.
Avere also acts as a “broker”, being a front-end for different Filers and offering replicas between those systems. If you replace a filer with a Cloud Storage, FXT can also be used as a cloud gateway.
The Avere FTX nodes offer feature like Tiering, Clustering between nodes, Replication (most used blocks are replicated several times in different systems to have more access points), Striping. These feateures, together with the different available tiers (RAM, SSD, SAS, SATA) inside every FXT machine, are responsibile for the performance boost that Avere offers. Also, Avere can be used both for file access, and also as a NFS storage for VMware, in fact they are VMware certified too.
When we talked about VMware, I reqested to move the discussion towards the “transparency” offered by Avere in regards to data stored into the backend NAS. The biggest problem of every caching system is in fact that data into the NAS are “behind” in he time line. If for example we configure a 4 hours cache, data inside the NAS itself are 4 hours old. This is not a problem during daily operations, but it could be a problem if for example we want to run a backup (or a snpshot) on the storage. Avere thought about this problem, and came out with a viable solution, even if I think it should be further improved. First you can configure the cache age, from 1 second 1 1 year; then you can configure some “clean periods”, when you want that your NAS is aligned with the cache. If you want to run a backup at 10 pm, you can schedule a cache flushing on Avere so that it can be clean by that time.
for the future, I do hope they will add an orchestration mechanism between Avere, client (maybe with also a VSS provider) and the NAS. In this way the whole process will be automatic, and cache would be flushed only when needed and not with a schedule.
Avere showed us some really interesting use cases of their “filer broker” solution. Avere nodes can be installed in a given location even if there is no local NAS there. Data are retrieved and managed by Avere directly via the WAN connection. The clear advantage compared to a WAN Accelerator is you do not need two systems at both ends. The remote site can be a headquarter, or a Cloud Storage provider.
Acting like a broker, Avere can leverage its “FlashMove” feature and copy data from one NAS to the other transparently, or between two Cloud Providers, or a mix. This feature allows for transparent migrations between two systems, for example when you are dismissing an old NAS with a new one, or if you are changing your Cloud Provider (think about the Nirvanix ”affaire”). Also, when the two NAS are in full sync, you can decide to detach the old NAS ot to keep both, and use another feature called “FlashMirror”.
With the upcoming realese numbered 4.0 of their AOS operating system, Avere will have even more “Cloud” support. There is already support for EMC Atmos (connected via NFS however, so just like any other local NAS), but the new version will support S3 libraries; this means it will be able to connect a Edge filer to Amazon Web Services, or a compatible solution like CleverSafe. Also OpenStack Swift support is coming, so the list of compatible Cloud Storage providers will increase.
There is also a specific support for AWS Glacier, but obviously the kind of interaction here is different, because of the nature of Glacier itself. Glacier does not support real time read/write operations, but it’s more an archival solution. So, within Avere you need to first “freeze” a directory, so it can be transferred to Glacier. This is a nice solution for all those situations (like closed projects) where you have huge amounts of data that you do not need to open anymore, but you only need to keep them online.
The use cases Avere showed us were really interesting, and they are another proof of an ongoing new concept that is raising lately: you should stop adding capacity to solve performance problems! You will be better thinking about a two layer design, even by using two different vendors if they can work together, just like Avere and a NAS solution. This will have a low price per GB, and will be responsible for capacity (Avere calls it Core), and another system with high performances and less disk space, responsible for performances (Edge).
Obviously, in the future some main vendors would eventually offer a NAS systems able to use NAND memories in a better way, but there are no upcoming solutions at the horizon, so Avere has a real advantage here. Avere has been designed specifically to address those problems, way better than retrofitting SSDs into a classic storage, never designed to use them.
Finally, the price of an Avere system is good: an entry level solution with 2 nodes has a list price around 50.000 USD, and can then be scaled only if and when needed. But even if you need to esign an environment from scratch rather than upgrading an existing one, an intriguing solution could be made with a some FXT system at the Edge and a bunch of really cheap barebone servers at the Core, with huge savings on your budget and with the same performances and redundancy levels, since these are guaranteed by the Avere frontend.