Grepular

The dedicated server which this website and all my other personal stuff runs on is hosted with PoundHost.com. They’re cheap, offer very good technical support, and have been generally reliable for the several years I’ve used them. They recently started offering a service called Data Storage. You can basically rent storage space by the gigabyte, and they’ll make it accessible to you via iSCSI. So if you have a dedicated server or colocated machine with them, you can mount that storage on it.

It’s cheap too. 17 pence per gigabyte for the cheapest option, and there are no per/bandwidth or per/IO charges like you’d receive with Amazons EC2 services (S3/EBS).

I started playing with it recently because I wanted to be able to share a filesystem between multiple servers for redundancy. I’ve used 100% free software. Specifically, I’ve been using Redhat Cluster Suite on a pair of Centos 5.3 VMs running on top of VMWare server, clustered LVM, and GFS. They use a HP LeftHand SAN, which supports SCSI reservations which means you can do proper automated Fencing) too.

All in all, it’s quite a neat solution. There are drawbacks though. Poundhost Dedicated Servers come with 100Mb NICs, which means you’re limited to a theoretical maximum of 12.5 megabytes per second to read/write from/to it. In practice, I get about 8 megabytes per second. I know this because I’ve been running hdparm against the device once every ten minutes for several days now and logging the results. Of course, I’m sharing the storage network and disks with other people, so my performance is going to be dictated by the behaviour of other users.

You can choose to pay twice as much per gigabyte for “two-way replicated” storage, and apparently it’s straight forward to change from one to the other. If you get two-way replicated storage then it has a 100% SLA, and any downtime means you get your full monthly cost refunded. I don’t have two-way replicated though, which means I get NO SLA. I didn’t realise this until yesterday. The storage stopped responding at about 11:30 in the morning, and it took them about four hours to get it back online. From following the status page, most of that time seemed to be in getting the HP engineer on site. That’s the trouble when you buy a proprietary solution, you can’t always fix it yourself. I’m not sure if it affected the two-way replicated storage, but it certainly affected mine.

I’m going to keep using it and monitoring its performance and uptime. It’s useful for backing up data, but I don’t currently trust it enough to run a live platform off it. Poundhost themselves obviously do, as they run their VPSs off it, so I expect downtime like this to be a rare thing.