Once a year (optionally twice), every Google employee completes a structured self-assessment about what they've done since the last performance review, what went well, and what should have been better. They then solicit feedback on this report from a handful of colleagues they've worked with. These colleagues are typically on their own teams, but not always. Generally speaking, comments from more senior employees are encouraged; pretty much everybody asks their tech leads to comment. Managers then write assessments of each employee's work based on the reports and peer feedback, producing a grade from (2.5, 3.7, ...) and prose comments.
The process itself isn't that unusual, but Google undertakes it in a very ponderous heavyweight manner. The peer feedback takes quite a bit of time to write, particularly for senior engineers. And Google insists that managers vet the grades they give through a complex series of calibration meetings up and down the company, with the goal being fair and uniform assessment for comparable positions. The grading is also very fine-grained. There can be a lot of discussion about whether someone is a 3.1 or a 3.3. The process takes nearly two months end-to-end, in part because it is combined with promotions, which are scrutinized even more closely. Engineers love to bitch about "perf", and the long-term trend is toward a simpler, lighter process.
My take on the matter is that Google tends to be very deliberate and bureaucratic in what it considers truly vital activities. Hiring, for example, is very slow and formalized. Checking in code can require a quest of approvals, tests, and automated verification steps. The performance review process is therefore slow and ponderous because Google wants to be very sure it is making the right decisions in performance assessment and promotion.